Download Evanescent Waves in the Near- and the Far Field

Evanescent Waves in the Near- and the Far Field
HENK F. ARNOLDUS
Department of Physics and Astronomy, Mississippi State University,
P.O. Drawer 5167, Mississippi State, Mississippi, 39762-5167, USA
[email protected]
Abstract. Radiation emitted by a localized source can be considered a combination of
traveling and evanescent waves, when represented by an angular spectrum. We have
studied both parts of the radiation field by means of the Green’s tensor for the electric
field and the “Green’s vector” for the magnetic field. It is shown that evanescent waves
can contribute to the far field, despite their exponential decay, in specific directions. We
have studied this far-field behavior by means of an asymptotic expansion with the radial
distance to the source as large parameter. As for the near field, we have shown explicitly
how the singular behavior of radiation in the vicinity of the source is entirely due to the
evanescent waves. In the process of studying the traveling and evanescent waves in both
the near field and the far field, we have found a host of new representations for the
functions that determine the Green’s tensor and the Green’s vector. We have obtained
new series involving Bessel functions, Taylor series, asymptotic series up to all orders
and new integral representations.
Advances in Imaging and Electron Physics 132 (2004) 1-67
Table of Contents
I. Introduction
4
II. Solution of Maxwell’s Equations
8
III. Green’s Tensor and Vector
10
IV. Electric Dipole
13
V. Angular Spectrum Representation of the Scalar Green’s Function
VI. Angular Spectrum Representation of the Green’s Tensor and Vector
VII. Traveling and Evanescent Waves
15
17
19
VIII. The Auxiliary Functions
22
IX. Relations Between the Auxiliary Functions
X. The Evanescent Part
25
26
XI. The Traveling Part
29
XII. The z-Axis
31
XIII. The xy-Plane
33
XIV. Relation to Lommel Functions
37
XV. Expansion in Series with Bessel Functions
XVI. Asymptotic Series
41
XVII. Evanescent Waves in the Far Field
XVIII.
38
43
Uniform Asymptotic Approximation
46
A. Derivation
47
B. Results
55
XIX. Traveling Waves in the Near Field
61
XX. The Coefficient Functions
63
2
XXI. Integral Representations
68
XXII. Evanescent Waves in the Near Field
71
XXIII. Integral Representations for the Evanescent Waves
74
XXIV. Conclusions
78
References
80
Appendix A
85
Appendix B
88
3
I. Introduction
Radiation emitted by a localized source of atomic dimensions is usually observed in the
far field with a macroscopic detector like a photomultiplier tube. This far-field wave is a
spherical wave and its modulation in phase and amplitude carries information on the
characteristics of the source. The recent dramatic advances in nanotechnology (Ohtsu,
1998) and the increasing experimental feasibility of measuring electromagnetic fields on
a length scale of an optical wavelength in the vicinity of the source with near-field
microscopes (Pohl, 1991; Pohl and Courjon, 1993; Courjon and Bainier, 1994; Paesler
and Moyer, 1996; Grattan and Meggitt, 2000; Courjon, 2003), has made it imperative to
study in detail the optical properties of radiation fields with a resolution of a wavelength
or less around the source.
It is well known that radiation emitted by a localized source has four typical
components, when considering the dependence on the radial distance to the source. Let
the source be located near the origin of coordinates, and let vector r represent a field
point. We shall assume that the radiation is monochromatic with angular frequency ω
with corresponding wave number ko = ω / c . The spherical wave in the far field
mentioned above then has an overall factor of exp(iko r ) / r , and this is multiplied by a
complex amplitude depending on the details of the source. The important property of this
component of the field is that it falls off with distance as 1 / r . Since the intensity is
determined by the square of the amplitude, the outward energy flow per unit area is
proportional to 1 / r 2 . When integrated over a sphere with radius r around the source, the
emitted power becomes independent of the radius of the sphere, and therefore can be
4
observed in “infinity”. Conversely, any component of the field that falls off faster than
1 / r will not contribute to the power at macroscopic distances. The complete radiation
field has three more components that become important when we want to consider optical
phenomena on a length scale of a wavelength. The field has a component proportional to
1 / r 2 , which is called the middle field, and a component that falls off as 1 / r 3 , which is
the near field contribution. In addition, there is a delta function in the field which only
exists inside the source, and this part is therefore usually omitted. It has been realized for
a long time that this delta function is necessary for mathematical consistency (Jackson,
1975), and more recently it appeared that for a proper account of the near field this
contribution can not be ignored any longer (Keller, 1996, 1999a, 1999b). Especially
when considering k-space descriptions of parts of the field, this delta function has to be
included, since it spreads out over all of k-space (Arnoldus, 2001, 2003a).
In near-field optics, a representation of the radiation field in configuration (r) space is
not always attractive, since all parts of the field (near, middle and far) contribute more or
less equally, depending on the distance to the source. Moreover, the separate parts are
not solutions of Maxwell’s equations individually, so the coupling between all has to be
retained. A description in configuration (k) space has the advantage that the Fourier
plane waves of the decomposition do not couple among each other, but the problem here
is that the separate plane waves do not satisfy Maxwell’s equations. The reason is that at
a given frequency ω , only plane waves with wave number ko = ω / c can be a solution of
Maxwell’s equations, whereas in k-space waves with any wave number k appear. The
solution to this problem is to adopt what is called an “angular spectrum” representation.
Here, we make a Fourier transform in x and y, but not in z, so this is a two-dimensional
5
Fourier transform in the xy-plane. The wave vector k | | in the xy-plane can have any
magnitude and direction. The idea is then that we associate with this k | | a threedimensional plane wave of the form exp(i K ⋅ r ) , with an appropriate complex amplitude,
such that this wave is a solution of Maxwell’s equations. In particular, the magnitude of
K has to be K = ko = ω / c , which implies that, given k | | , the z-component of K is fixed,
apart from a possible minus sign. The wave vector K of a partial wave in this
representation is given by
K = k | | + β sgn( z ) e z
(1)
with sgn(z ) the sign of z, and with the parameter β defined as


β =
i

ko2 − k|2|
, k || < ko
k|2|
, k || > ko
− ko2
.
(2)
For k || < ko , this parameter is positive, and therefore the sign of K z is the same as the
sign of z. Hence, exp(i K ⋅ r ) is a traveling plane wave, which travels in the direction
away from the xy-plane. For k || > ko , β is positive imaginary, and this corresponds to a
wave which decays in the z-direction. The sign of K z is chosen such that for z > 0 the
wave decays in the positive z-direction and for z < 0 it decays in the negative z-direction,
a choice which is obviously dictated by causality. Waves of this type are called
evanescent waves, and in an angular spectrum representation these waves have to be
included. Since k | | is real, the evanescent waves travel along the xy-plane in the
direction of k | | . Figure 1 schematically illustrates the two types of waves in the
angular spectrum.
6
The advantage of the fact that each partial wave in the angular spectrum is a solution
of the free-space Maxwell equations can not be overemphasized. For instance, when the
source is located near an interface, each partial wave reflects and refracts in the usual
way, and this can be accounted for by Fresnel reflection and transmission coefficients.
The total reflected and refracted fields are then simply superpositions of these partial
waves, and the result is again an angular spectrum representation. This yields an exact
solution for the radiation field of a source near an interface (Sipe, 1981, 1987), and the
result has been applied to calculate the radiation pattern of a dipole near a dielectric
interface (Lukosz and Kunz, 1977a, 1977b; Arnoldus and Foley, 2003b, 2003d) and a
nonlinear medium (Arnoldus and George, 1991), and to the computation of the lifetime
of atomic states near a metal (Ford and Weber, 1981, 1984).
Evanescent waves have a long history, going back to Newton (de Fornel, 2001), and
common wisdom tells us that evanescent waves dominate the near field whereas the
traveling waves in the angular spectrum account for the far field. The latter statement
derives from the fact that evanescent waves die out exponentially, away from the xyplane, and can therefore not contribute to the far field. On the other hand, near the source
each traveling and each evanescent wave in the angular spectrum is finite in amplitude,
giving no obvious reason why evanescent waves are more prominent in the near field
than traveling waves. In this paper we shall show explicitly that in particular the
singularity of the field near the origin (as in 1 / r 3 , etc.) results entirely from the
contribution of the evanescent waves. On the other hand, we shall show that evanescent
waves do end up in the far field, despite their exponential decay, defying common sense.
7
II. Solution of Maxwell’s Equations
We shall consider a localized source of radiation in which the charge density ρ (r, t ) and
the current density j(r, t ) oscillate harmonically with angular frequency ω . We write
j(r, t ) = Re[ j(r ) e −iω t ]
(3)
with j(r ) the complex amplitude, and similarly for ρ (r, t ) . The electric field E(r, t ) and
the magnetic field B(r, t ) will then have the same time dependence, and their complex
amplitudes are E(r ) and B(r ) , respectively. We assume the charge and current densities
to be given (as is for instance the case for a molecule in a laser beam). The electric and
magnetic fields are then the solution of Maxwell’s equations:
∇ ⋅ E(r ) = ρ (r ) / ε o
(4)
∇ × E(r ) = iω B(r )
(5)
∇ ⋅ B(r ) = 0
(6)
iω
∇ × B(r ) = − 2 E(r ) + µ o j(r ) .
c
(7)
If we take the divergence of Eq. (7) and use Eq. (4) we find
∇ ⋅ j(r ) = iωρ (r )
(8)
which expresses conservation of charge.
In order to obtain a convenient form of the general solution we temporarily introduce
the quantity
P(r ) =
iωµo
4π
∫ d r' j(r' ) g (r − r' )
3
8
(9)
where g (r ) is the Green’s function of the scalar Helmholz equation
g (r ) =
eiko r
.
r
(10)
It follows by differentiation that
∇ 2 g (r − r ' ) = −ko2 g (r − r ' ) , r ' ≠ r .
(11)
If we then wish to evaluate ∇ 2P(r ) , then it seems that the entire r dependence enters
through g (r − r ' ) in the integrand, and with Eq. (11) this would give ∇ 2P(r ) = − ko2P(r ) .
However, it should be noted that g (r ) has a singularity at r = 0 , and therefore the
integrand of the integral in Eq. (9) is singular at r ' = r . When the field point r is inside
the source, it is understood that a small sphere around r is excluded from the range of
integration. When we vary r, by applying the operator ∇ 2 on P(r ) , then we also move
the small sphere. It can then be shown (van Kranendonk and Sipe, 1977; Born and Wolf,
1980) that this leads to an extra term when moving ∇ 2 under the integral. The result is
∇ 2P(r ) = − ko2P(r ) − iωµo j(r ) .
(12)
It can then be verified by inspection that the solution of Maxwell’s equations is
1
E(r ) = P(r ) + 2 ∇(∇ ⋅ P(r ))
ko
−i
B(r ) =
∇ × P(r )
ω
taking into consideration relation (8) between the charge and current densities.
9
(13)
(14)
In Eq. (13), the operator ∇(∇ ⋅ ...) acts on the integral in Eq. (9), and when we move
this operator under the integral an additional term appears, similar to the second term on
the right-hand side of Eq. (12). In this case we find
E(r ) = −
i
3ε o ω
j(r ) +
i
4π ε o
d r' ( k
ω∫
3
2
o
j(r ' ) g (r − r ' )
)
+ ∇{∇ ⋅ [ j(r ' ) g (r − r ' )]} .
(15)
The ∇(∇ ⋅ ...) in the integrand only acts on the r dependence of g (r − r ' ) , and therefore
this can be written as
∇{∇ ⋅ [ j(r ' ) g (r − r ' )]} = [ j(r ' ) ⋅ ∇]∇g (r − r ' ) .
(16)
For the magnetic field we have to move ∇ × ... under the integral, but this does not lead to
an additional term. We thus obtain
B(r ) =
−1
4π ε o c
∫ d r' j(r' )× ∇g (r − r' )
3
2
(17)
where we have used ∇ × [ j(r ' ) g (r − r ' )] = − j(r ' ) × ∇g (r − r ' ) .
III. Green’s Tensor and Vector
The solutions for E(r ) and B(r ) from the previous section can be cast in a more
transparent form by adopting tensor notation. To this end we notice that the right-hand
side of Eq. (16) can be written as
[ j(r ' ) ⋅ ∇]∇g (r − r ' ) = [∇∇g (r − r ' )] ⋅ j(r ' ) .
10
(18)
Here, ∇∇g (r − r ' ) is a tensor with a dyadic structure (given below), and the dot product
between a dyadic form ab and a vector c is defined as (ab) ⋅ c = a(b ⋅ c) in terms of the
regular dot product between the vectors b and c. The result is a vector proportional to a.
t
t
The unit tensor I has the effect of I ⋅ a = a . The solution (15) for E(r ) can then be
written as
E(r ) = −
i
3ε o ω
j(r ) +
i
4π ε o ω
∫
t
d 3r ' [ko2 I g (r − r ' )
+ ∇∇g (r − r ' )] ⋅ j(r ' ) .
(19)
In order to simplify this even more, we write the current density in the first term on the
right-hand side as
t
∫
j(r ) = d 3r ' [ I δ (r − r ' )] ⋅ j(r ' )
(20)
and then we combine the two integrals. The solution then takes the compact form
E(r ) =
iko2
4π ε o ω
∫
t
d 3r ' g (r − r ' ) ⋅ j(r ' ) .
(21)
t
Here, g (r ) is the Green’s tensor, defined as
tg (r) = − 4π δ (r)tI +  tI +


3ko2

 g (r ) .
∇∇

ko2

1
This tensor has been studied extensively, and a book (Tai, 1971) is devoted to its use,
although, oddly enough, the delta function on the right-hand side was not included.
11
(22)
t
In order to find g (r ) explicitly, we only need to work out the derivatives ∇∇g (r ) .
At this point it is convenient to adopt dimensionless variables for coordinates with 1 / ko
as unit of measurement. The dimensionless vector representing the field point will be
denoted by q = kor . The magnitude of this vector, q = ko r , is then the dimensionless
distance of the field point from the origin, and such that q = 2π corresponds to a distance
of one optical wavelength. We shall also introduce the dimensionless Green’s tensor by
tχ (q) = 1 tg (r) .
(23)
ko
This Green’s tensor is then found to be
tχ (q) = − 4π δ (q)tI + (tI − 3qˆ qˆ )(i − 1 ) eiq + (tI − qˆ qˆ ) eiq
q q2
3
q
(24)
from Eq. (22). The radial unit vector q̂ is the same as r̂ , and δ (q) = δ (r ) / ko3 is the
dimensionless delta function. The final expression for the electric field of a localized
source then becomes
E(r ) =
iko3
4π ε o ω
∫
t
d 3r ' χ (ko (r − r ' )) ⋅ j(r ' ) .
(25)
The result for the magnetic field can be rewritten in a similar way, but this is a lot
simpler. We define the dimensionless vector quantity
1
η (q) = − 2 ∇g (r )
ko
in terms of which the magnetic field becomes
12
(26)
B(r ) = −
ko3
d r ' η(k (r − r ' )) × j(r ' )
4π ε ω c ∫
3
o
o
(27)
greatly resembling Eq. (25) for the electric field. Apparently, the vector η(q) plays the
same role for the magnetic field as the Green’s tensor for the electric field, although it
should be noted that this vector is not a Green’s function in the usual sense. This Green’s
vector has the explicit form
 1  eiq
η(q) =  − i qˆ
.
q  q
(28)
IV. Electric Dipole
We now consider a localized charge and current distribution of the most important form:
the electric dipole. Its importance comes from the fact that most atomic and molecular
radiation is electric dipole radiation. To see how this limit arises, we first consider a
general distribution. Let the material be made up of particles, numbered with the
subscript α. Each particle has a position vector rα (t ) , velocity
vα (t ) =
d
rα (t )
dt
(29)
and electric charge qα . The dipole electric dipole moment d(t ) of the distribution is
defined as
d(t ) =
qα rα (t ) .
∑
α
13
(30)
The time dependent current density can be expressed as (Cohen-Tannoudji et.al., 1989)
j(r, t ) =
qα vα (t )δ (r − rα (t )) .
∑
α
(31)
We now assume that the linear dimensions of the distribution are very small, and
centered around a given point ro . We then have rα (t ) ≈ ro , and Eq. (31) becomes
j(r, t ) = δ (r − ro )
qα vα (t ) .
∑
α
(32)
d
d(t) .
dt
(33)
Comparison with Eq. (29) gives
j(r, t ) = δ (r − ro )
Since the current distribution completely determines the electric and magnetic fields,
according to Eqs. (25) and (27), we simply define an electric dipole, located at ro , as a
distribution with j(r, t ) given by Eq. (33).
For time harmonic fields, the dipole moment has the form
d(t ) = Re[d e −iω t ]
(34)
where d is an arbitrary complex-valued vector. The current density follows from Eq.
(33), and comparison with Eq. (3) then gives for the time-independent current density
j(r ) = − iω δ (r − ro )d .
(35)
The corresponding charge density follows from Eq. (8):
ρ (r ) = − d ⋅ ∇δ (r − ro )
14
(36)
although we don’t need that for the present problem.
Due to the delta function in Eq. (35), the integrals in Eqs. (25) and (27) can be
evaluated. For a dipole located at the origin of coordinates we obtain for the fields
E(r ) =
B(r ) =
ko3
4π ε o
tχ (q) ⋅ d
i ko3
η(q) × d
c 4π ε o
(37)
(38)
with q = kor . This very elegant result shows that the spatial dependences of the Green’s
tensor and vector are essentially the spatial distribution of the electric and magnetic field
of dipole radiation (apart from the tensor and cross product with d).
The composition of the electric field now follows from Eq. (24). The first term on the
right-hand side is a delta function, which only exists in the dipole. We call this the self
field. The second term has a 1 / q3 and a 1 / q 2 contribution, which are the near and the
middle field, respectively. The last term falls off as 1 / q , and this is the far field.
Similarly, for the magnetic field we see from Eq. (28) that this field only has a far and a
middle field, but no near or self field.
V. Angular Spectrum Representation of the Scalar Green’s Function
As mentioned in the Introduction, for many applications the representation of the Green’s
tensor and vector as in Eqs. (24) and (28), respectively, is not practical. In this section
we shall first consider the scalar Green’s function, given by Eq. (10). In order to derive a
more useful representation, we first transform to k-space. The transformation is
15
∫
3 e
G (k ) = d r
iko r
r
e −ik ⋅r .
(39)
For a given k, we use spherical coordinates and such that the z-axis is along the k vector.
First we integrate over the angles. Then the remaining integral over the radial distance
does formally not exist, and we have to include a small positive imaginary part iε in ko .
The resulting integral can be evaluated with contour integration, and the result is
4π
G (k ) = − 2
, ε ↓0 .
k o − k 2 + iε
(40)
The inverse is then
g (r ) = −
1
1
3
d
k
ei k ⋅r , ε ↓ 0 .
2
2
2
k o − k + iε
2π
∫
(41)
This integral can be calculated by using spherical coordinates in k-space. The result is
again exp(iko r ) / r , which justifies the construction with the small imaginary part in the
wave number.
Instead of using spherical coordinates in k-space, we now consider Cartesian
coordinates for the integral in Eq. (41). With contour integration we perform the integral
over k z , which yields
g (r ) =
i
2π
∫
d 2k ||
1 i K ⋅r
e
.
β
(42)
The parameter β is defined in Eq. (2) and the wave vector K is given by Eq. (1). The
integral runs over the entire k || plane, which is the xy-plane of k-space. Equation (42) is
the celebrated angular spectrum representation of the scalar Green’s function.
16
As explained in the Introduction, Eq. (42) is a superposition of traveling and
evanescent waves. Inside the circle k || = ko in the k || plane, the waves exp(iK ⋅ r ) are
traveling because β , and thereby K z , is real, and outside the circle these waves are
evanescent since their wave vectors have an imaginary z-component. Just on the circle
we have β = 0 , and the integrand has a singularity. We shall see below that this
singularity is integrable and poses no problems.
VI. Angular Spectrum Representation of the Green’s Tensor and Vector
t
In order to find an angular spectrum representation for the Green’s tensor g (r ) , it would
be tempting to take expression (42) for the scalar Green’s function, substitute this in the
right-hand side of Eq. (22), and then move the operator in large brackets under the k ||
integral. This procedure leads to the wrong result in that it misrepresents the self field
(the delta function on the right-hand side of Eq. (22)). The delta function in the Green’s
tensor came from moving the ∇(∇ ⋅ ...) operator in Eq. (13) under the integral sign in Eq.
(9) for P(r ) , and this led to Eq. (15). The extra term came from the singularity at r = 0
of g (r ) = exp(iko r ) / r in P(r ) . When we represent g (r ) by its angular spectrum, Eq.
(42), substitute this in Eq. (9) for P(r ) , and change the order of integration, we obtain
P(r ) = −
ωµo
1
d 2k | |
2
β
8π
∫
∫ d r' j(r' )e
3
iK ( z − z ')⋅(r −r ')
.
Here we have shown explicitly the z-dependence of K(z). If we now consider the
operator ∇(∇ ⋅ ...) acting on P(r ) , then the singularity at r ' = r has disappeared.
17
(43)
Therefore, when we change the order of integration, the action of ∇(∇ ⋅ ...) does not
“move the sphere” anymore, and we can freely move this operator under the r ' integral.
t
All following steps are the same, leading to Eq. (22) for g (r ) . Therefore, we can
substitute the angular spectrum representation (42) into Eq. (22) and move the derivatives
under the integral, but we have to leave out the delta function on the right-hand side of
Eq. (22). This then yields
tg (r) =
i
2π
∫
d 2k ||
t

1 
1
I + 2 ∇∇  ei K ⋅r .

β 
ko

(44)
The dyadic operator ∇∇ now only acts on the exponential exp(iK ⋅ r ) , and we can take
the derivatives easily. Care should be exercised, however, since K depends on z through
sgn( z ) . With
d
sgn( z ) = 2δ ( z )
dz
(45)
∇∇eiK ⋅r = [2iβδ ( z )e z e z − KK ] eiK ⋅r .
(46)
we find
Furthermore we use the spectral representation of the two-dimensional delta function
∫d k
2
|| e
ik || ⋅r
= 4π 2δ ( x)δ ( y )
(47)
and when we then put everything together we obtain the angular spectrum representation
of the dimensionless Green’s tensor:
18
tχ (q) = − 4π δ (q)e e
z z
+
i
2π ko
∫
d 2k ||
t

1 
1
I − 2 KK  eiK ⋅r .

β 
ko

(48)
It is interesting to notice that a new delta function appears on the right-hand side, which
comes from the discontinuous behavior of K ( z ) at z = 0 . When compared to the
representation (24) or (22) in r-space, we see that here we have a different delta function.
Since the previous one represented the self field, the delta function in Eq. (48) must be
something different. We will get back to this point in Sec. 8 and Appendix A.
The Green’s vector for the magnetic field does not have any of these complications,
and from Eq. (26) we immediately obtain
η(q) =
1
∫d k
2
2π ko2
1
||
β
K eiK ⋅r
(49)
since ∇ exp(iK ⋅ r ) = iK exp(iK ⋅ r ) .
VII. Traveling and Evanescent Waves
t
The Green’s tensor χ (q) in Eq. (48) splits naturally into three parts:
tχ (q) = − 4π δ (q)e e + tχ (q)tr + tχ (q)ev .
z z
t
(50)
t
Here, χ (q)tr is the part of χ (q) which only contains the traveling waves, e.g.,
tχ (q)tr =
i
2π ko
∫
d 2k ||
k|| < k o
19
t

1 
1
I −
KK  eiK ⋅r

β 
ko2

(51)
t
where the integration only runs over the inside of the circle k|| = ko . Similarly, χ (q)ev is
the part which only contains the evanescent waves. The Green’s vector η(q) for the
magnetic field has two parts
η(q) = η(q)tr + η(q)ev
(52)
in obvious notation.
We shall use both spherical coordinates (r ,θ ,φ ) and cylinder coordinates ( ρ ,φ , z )
for a field point, and most of the time we shall use the dimensionless coordinates
q = ko r , ρ = ko ρ and z = ko z . The radial unit vector in the xy-plane is given by e ρ =
e x cos φ + e y sin φ , in terms of which we have q = ρ e ρ + ze z , and the tangential unit
vector is eφ = − e x sin φ + e y cos φ . The relation to spherical coordinates is ρ = q sin θ ,
z = q cosθ , from which qˆ = sin θ e ρ + cosθ e z . Let us now consider the integration over
the k || -plane. For a given field point r, we take the direction of the ~
x -axis in the k || ~
plane along the corresponding e ρ , and we measure the angle φ from this axis, as shown
in Fig. 2. The dimensionless magnitude of k || will be denoted by α = k || / ko , which
implies that the range 0 ≤ α < 1 represents traveling waves and the range 1 < α < ∞
represents evanescent waves. We further introduce
β =
β
ko
= 1−α 2
(53)
with the understanding that β is positive imaginary for α > 1 , as in Eq. (2). From Fig. 2
~
~
we see that k || = α ko (e ρ cos φ + eφ sin φ ) and therefore
~
~
K = α ko (e ρ cos φ + eφ sin φ ) + ko β sgn( z )e z
20
(54)
~
from which we find K ⋅ r = αρ cos φ + β | z | . Here we used sgn( z ) z = | z | . Combining
everything then gives the following translation for an integral over the k || -plane
∫
∞
1
α
d 2k || eiK ⋅r (...) = ko dα eiβ | z |
∫0
β
β
2π
∫0
~
~
dφ eiαρ cosφ (...)
(55)
where the ellipses denote an arbitrary function.
Let us now consider the traveling part of the Green’s tensor at the origin of
coordinates. We set r = 0 in Eq. (51) and use representation (55) for the k || integral.
~
The only dependence on φ comes in through KK, with K given by Eq. (54), and the
~
integral over φ can be performed directly. For the remaining integral over α we make a
change of variables according to u = (1 − α 2 )1 / 2 , after which the integral over u is
elementary. Furthermore we recall the resolution of the unit tensor in cylinder
coordinates
t
I = e ρ e ρ + eφ eφ + e z e z
(56)
tχ (0)tr = 2 i tI .
(57)
which then gives
3
The most important conclusion of this simple result is that the traveling part of the
Green’s tensor is finite at the origin. Since the Green’s tensor itself is highly singular at
this point, we conclude that any singularity at q = 0 must come from the evanescent
waves. This also justifies the opinion that near the origin the field of a dipole (or the
21
Green’s tensor) is dominated by the evanescent waves. In the same way we obtain for
the Green’s vector
1
η(0)tr = sgn( z )e z
2
(58)
which is also finite.
VIII. The Auxiliary Functions
In order to study the behavior of the traveling and evanescent waves in detail, we go back
to Eq. (48), and we write the k || integral as in Eq. (55). Since we now have the
~
~
exponential of K ⋅ r = αρ cos φ + β | z | , the integrals over φ lead to Bessel functions, as
for instance
2π
∫
~
~
dφ eiαρ cos φ = 2π J 0 (αρ ) .
(59)
0
After some rearrangements, the Green’s tensor then takes the form
tχ (q) = − 4π δ (q)e e
t
1
1
z z + 2 ( I + e z e z ) M a (q) + 2 (eφ eφ − e ρ e ρ ) M b (q)
t
+ 12 sgn( z )(e ρ e z + e z e ρ ) M c (q) + 12 ( I − 3e z e z ) M d (q)
(60)
where we have introduced four auxiliary functions
∞
∫0
M a (q) = i dα
α
J 0 (αρ ) eiβ | z |
β
22
(61)
∞
∫0
M b (q) = − i dα
α3
J 2 (αρ ) eiβ | z |
β
(62)
∞
∫0
M c (q) = 2 dα α 2 J1 (αρ ) eiβ | z |
(63)
∞
∫0
M d (q) = i dα αβ J 0 (αρ ) eiβ | z | .
(64)
These functions are functions of the field point q. They depend on the cylinder
coordinates ρ and z but not on φ . With ρ = q sin θ , z = q cosθ , we can also see them
as functions of the spherical coordinates q and θ . Furthermore, the z -dependence only
enters as | z | , and therefore these functions are invariant under reflection in the xy-plane.
For α > 1 , β is positive imaginary, and exp(iβ | z |) decays exponentially with | z | ,
which guarantees the convergence of the integrals. The exception is z = 0 for which
some of the integrals do not exist in the upper limit. We know, however, that the Green’s
tensor is finite for all points in the xy-plane except the origin, so the limit z → 0 has to
exist. The factors in front of the functions in Eq. (60) show explicitly the tensorial part of
the tensor.
In the same way the Green’s vector can be written as
η(q) = sgn( z ) e z M e (q) + eρ M f (q)
(65)
which involves two more auxiliary functions
∞
∫0
M e (q) = dα α J 0 (αρ ) eiβ | z |
23
(66)
∞
∫0
M f (q) = i dα
α2
J1(αρ ) eiβ | z | .
β
(67)
We now have expression (60) for the Green’s tensor and expression (24), and these
must obviously be the same. We set qˆ = sin θ e ρ + cosθ e z in Eq. (24) and compare to
Eq. (60). When equating the corresponding tensorial parts we obtain four equations for
the four auxiliary functions. Upon solving this yields the explicit forms
eiq
M a (q) =
q
(68)
 3  1  eiq
M b (q) = sin θ 1 +  i − 
 q  q  q
(69)
2

M c (q) = − | sin 2θ | 1 +

3  1  eiq
 i − 
q  q  q
 1  eiq
8π
3  1  eiq
2 
M d (q) = −
δ (q) −  i −  2 + cos θ 1 +  i − 
.
3
 q q
 q  q  q
(70)
(71)
We see that M a (q) is the scalar Green’s function from Eq. (10), apart from a factor of
ko , but the other three are more complicated. In particular, M d (q) has a delta function,
which, when added to the delta function in Eq. (60), gives exactly the self field part in
Eq. (24). In Appendix A we show that the integral representation (64) contains indeed a
delta function, and that it resides entirely in the evanescent part.
Similarly, comparison of Eqs. (28) and (65) gives
 1  eiq
M e (q) = | cosθ |  − i 
q  q
24
(72)
 1  eiq
M f (q) = sin θ  − i 
.
q  q
(73)
IX. Relations Between The Auxiliary Functions
From Eqs. (69) and (70) we observe the relation
M c (q) = − 2
|z|
ρ
M b (q)
(74)
since | cosθ | / sin θ = | z | / ρ . Less obvious is:
M d (q) = M a (q) − M b (q) −
2
ρ
M f (q)
(75)
as will be shown in Appendix A. Another relation that we notice immediately is
M e (q) =
|z|
ρ
M f (q) .
(76)
Then Eq. (65) becomes
η(q) =
1
ρ
q M f (q)
(77)
as interesting alternative.
When we differentiate the integral representation (61) with respect to z and use
d
| z | = sgn( z )
dz
we obtain the relation
25
(78)
∂
M a (q) = − sgn( z ) M e (q)
∂z
(79)
∂
M e (q) = sgn( z ) M d (q)
∂z
(80)
∂
M f (q) = − 12 sgn( z ) M c (q) .
∂z
(81)
and similarly
We can also differentiate with respect to ρ . With J 0 ( x)' = − J1( x) we find from Eq. (61)
∂
M a (q) = − M f (q)
∂ρ
(82)
∂
∂
M e (q) = sgn( z )
M f (q) .
∂ρ
∂z
(83)
and similarly
Many other relations can be derived, especially involving higher derivatives.
X. The Evanescent Part
t
The evanescent part χ (q)ev of the Green’s tensor is given by Eq. (51), except that the
integration range is k || > ko . When expressed in auxiliary functions it becomes
tχ (q)ev = 1 (tI + e e
2
z z ) M a (q)
ev
+ 12 (eφ eφ − e ρ e ρ ) M b (q)ev
t
+ 12 sgn( z )(e ρ e z + e z e ρ ) M c (q)ev + 12 ( I − 3 e z e z ) M d (q)ev
26
(84)
and the functions M k (q)ev , k = a,b,... are the evanescent parts of the functions defined
by integral representations in Sec. VIII. This simply means that the lower integration
limits become α = 1 . The evanescent part of the Green’s vector η(q) is defined
similarly.
For α > 1 the parameter β is positive imaginary: β = i (α 2 − 1)1 / 2 . The following
theorem for n = 0, 1, ...
∞
∫1
dα
α n +1
2
1
J n (αρ ) e −| z | α −1 =
Jn (ρ )
|z|
α 2 −1
+
ρ
|z|
∞
∫1
2
dα α n J n −1(αρ ) e −| z | α −1
(85)
can be proved as follows. In the integral on the left-hand side, substitute the identity
α
2
1 d −| z | α 2 −1
e −| z | α −1 = −
e
2
|
|
α
z
d
α −1
(86)
and integrate by parts. For the derivative in the integrand of the remaining integral use
( x n J n ( x))' = x n J n −1( x) .
For n = 2 , Eq. (85) can be written as
M c (q)ev = −
2
ρ
(J 2 ( ρ ) + | z | M b (q)ev )
(87)
(J1( ρ ) − | z | M f (q)ev ) .
(88)
and for n = 1 it becomes
M e (q)ev = −
1
ρ
27
It is interesting to notice the similarity to Eqs. (74) and (76), respectively. Due to the
splitting, an additional Bessel function appears on the right-hand side. In Appendix A it
is shown that
M d (q)ev = M a (q)ev − M b (q)ev −
2
ρ
M f (q)ev
(89)
which is identical in form to Eq. (75) for the unsplit functions. Here, no additional term
appears. Eqs. (87)-(89) show that if we are able to compute the evanescent parts of
M a (q) , M b (q) and M f (q) then we also know the evanescent parts of the other three.
Since we also know the sum of the traveling and evanescent parts, we will also know the
traveling parts of the auxiliary functions. Also the relations involving derivates in Sec.
IX carry over to traveling and evanescent waves, since these were derived from the
integral representations without using explicitly the limits of integration.
In the integral representations of the evanescent parts of the auxiliary functions we
make the change of variables u = (α 2 − 1)1 / 2 , which leads to the new representations
∞
M a (q)
∫0
= du J 0 ( ρ 1 + u 2 ) e −u| z |
ev
(90)
∞
M b (q)
ev
M c (q)
ev
∫0
= − du (1 + u 2 ) J 2 ( ρ 1 + u 2 ) e −u| z |
(91)
∞
∫0
= 2 du u 1 + u 2 J1( ρ 1 + u 2 ) e −u| z |
(92)
∞
M d (q)
ev
∫0
= − du u 2 J 0 ( ρ 1 + u 2 ) e −u| z |
28
(93)
∞
M e (q)
ev
∫0
= du u J 0 ( ρ 1 + u 2 ) e −u| z |
(94)
∞
∫0
M f (q)ev = du 1 + u 2 J1 ( ρ 1 + u 2 ) e −u| z | .
(95)
As a first observation we notice that the singularities (the factors 1 / β ) in the lower limit
have disappeared, which proves that these singularities are indeed integrable. A second
point to notice is that the evanescent parts are pure real. Conversely, this means that the
entire imaginary parts of the Green’s tensor and vector consist of traveling waves. Or,
only the real parts of the auxiliary functions split
Re M k (q) = Re M k (q)tr + M k (q)ev , k = a, b,... f .
(96)
Since we know M k (q) , given by Eqs. (68)-(73), we also know the real parts. Therefore,
if we know either the first or the second term on the right-hand side of Eq. (96), we know
the other. We shall make use of this frequently.
XI. The Traveling Part
The traveling parts of the auxiliary functions, M k (q)tr , are given by the integral
representations of Sec. VIII with the integrations limited to 0 ≤ α < 1 and the
corresponding Green’s tensor then follows from Eq. (84) with the superscripts ev
replaced by tr. As explained in the previous section, the imaginary parts of these
functions are
Im M k (q)tr = Im M k (q) , k = a, b,... f
29
(97)
and the functions on the right-hand side are the imaginary parts of the right-hand sides of
Eqs. (68)-(73). So we shall only be concerned with the real parts of the functions
M k (q)tr .
We now make the change of variables u = (1 − α 2 )1 / 2 in the integral representations,
and we take the real parts. This yields the following representations
1
∫0
tr
Re M a (q) = − du J 0 ( ρ 1 − u 2 ) sin(u | z |)
(98)
1
∫0
Re M b (q)tr = du (1 − u 2 ) J 2 ( ρ 1 − u 2 ) sin(u | z |)
(99)
1
tr
∫0
Re M c (q) = 2 du u 1 − u 2 J1( ρ 1 − u 2 ) cos(u | z |)
(100)
1
∫0
Re M d (q)tr = − du u 2 J 0 ( ρ 1 − u 2 ) sin(u | z |)
(101)
1
∫0
tr
Re M e (q) = du uJ 0 ( ρ 1 − u 2 ) cos(u | z |)
(102)
1
tr
∫0
Re M f (q) = − du 1 − u 2 J1( ρ 1 − u 2 ) sin(u | z |) .
(103)
We shall use these representations for numerical integration, and in the graphs of the
following sections, this will be referred to as the “exact” solutions. By computing these
integrals numerically, we also have a reference for the evanescent parts, according to Eq.
(96). It should be noted that for large values of ρ and/or | z | this becomes very
computer time consuming due to the fast oscillations of the integrands.
30
Equations (74)-(76) show relations between the unsplit functions. If we take the real
parts of these equations, they still hold in the same form since all terms are real. The
corresponding relations for the traveling parts then follow by taking the difference with
Eqs. (87)-(89) for the evanescent parts, according to Eq. (96). We thus find
(
J 2 ( ρ ) − | z | Re M b (q)tr )
ρ
1
Re M e (q)tr = (J1( ρ ) + | z | Re M f (q)tr )
ρ
Re M c (q)tr =
2
Re M d (q)tr = Re M a (q)tr − Re M b (q)tr −
2
ρ
Re M f (q)tr .
(104)
(105)
(106)
XII. The z-Axis
Let us consider a field point on the z-axis ( z ≠ 0 ). We then have ρ = 0 , and with
J 0 (0) = 1 we find from Eq. (90)
M a (q)ev =
1
|z|
(107)
for the scalar Green’s function. Similarly, from Eqs. (93) and (94) we obtain
M d (q)ev = −
M e (q)ev =
2
| z |3
1
| z |2
.
(108)
(109)
Since J n (0) = 0 for n ≠ 0 the remaining functions vanish on the z-axis:
M b (q)ev = M c (q)ev = M f (q)ev = 0 .
31
(110)
The evanescent parts of the Green’s tensor and vector then become
tχ (q)ev = 1 (tI + e e
2
z z)
t
1
1
− ( I − 3e ze z ) 3
|z|
|z|
η(q)ev = sgn( z ) e z
1
| z |2
(111)
.
(112)
On the z-axis, we have | z | = q so the first term on the right-hand side of Eq. (111) is of
the far field type, being O ( 1/q), or O ( 1/r). As mentioned in the Introduction, fields that
drop off with distance as O ( 1/r) can be detected at a macroscopic distance from the
source. It seems counterintuitive that waves which decay exponentially in the z-direction
can survive in the far field on this z-axis. We also notice that the Green’s vector,
representing the magnetic field, is O (1/q2). We thus conclude that the electric evanescent
waves end up in the far field on the z-axis, but the corresponding magnetic evanescent
waves do not.
Some years ago, the subject of evanescent waves in the far field of an electric dipole
was vigorously debated in the literature. The origin of the controversy goes back to a
series of papers by Xiao (for instance, Xiao, 1996), who also derived Eq. (111) for the
Green’s tensor on the z-axis. He made the unfortunate mistake to conclude that since the
t
z-axis is an arbitrary axis in space, Eq. (111) should hold for all directions, so χ (q)ev for
all r should follow from Eq. (111) by replacing | z | by q and e z by q̂ (in our notation).
Wolf and Foley (1998) responded by noting that evanescent waves can only contribute to
the far field along the z-axis (or the xy-plane), similar to the Stokes phenomenon in
asymptotic analysis, and that this whole issue is of no interest and just a mathematical
32
oddity. We shall see below that this conclusion is also incorrect, although closer to the
truth. This discussion continued for a while (Xiao, 1999; Carney, et.al., 2000; Lakhtakia
and Weiglhofer, 2000; Xiao, 2000) until the correct solution to this problem was
presented by Shchegrov and Carney (1999) and Setälä, et.al. (1999).
On the z-axis we have qˆ = sgn( z ) e z , q = | z | and the unsplit Green’s tensor and
vector follow from Eqs. (24) and (28), respectively:
tχ (q) = (tI − e e
z z)
t

ei | z |
1  ei | z |
+ ( I − 3 e z e z ) i −

|z|
 | z |  | z |2
 ei| z |
 1
η(q) = sgn( z ) e z 
.
− i 
| z |  | z |
(113)
(114)
It should be noted that the tensor structure in Eq. (113) is different from the structure in
Eq. (111). Another noticeable difference is that the evanescent tensor and vector do not
have the factors exp(i | z |) , and therefore do not correspond to outgoing spherical waves.
XIII. The xy-Plane
Next we consider the situation in the xy-plane. Here we have z = 0 , and the integrals
defining the evanescent parts of the auxiliary functions, Eqs. (90)-(95), may not exist. To
get around this we first consider the traveling part. From Eqs. (98), (99), (101) and (103)
we obtain
Re M a (q)tr = Re M b (q)tr = Re M d (q)tr = Re M f (q)tr = 0
33
(115)
because sin(u | z |) = 0. The two remaining ones involve integrals over Bessel functions,
but with Eqs. (104) and (105) we immediately obtain
Re M c (q)tr =
Re M e (q)tr =
2
ρ
1
ρ
J2(ρ )
(116)
J1( ρ ) .
(117)
On the other hand, the real parts of M k (q) follow by taking the real parts of the righthand sides of Eqs. (68)-(73), after which the M k (q)ev ’s follow by taking the difference
with Eqs. (115)-(117), according to Eq. (96). We find
M a (q)ev =
ρ
3 
cos ρ 
 sin ρ +

2
ρ
ρ 
ρ 
2
M c (q)ev = − J 2 ( ρ )
M b (q)ev =
cos ρ
cos ρ
−
ρ
1 
cos ρ 
 sin ρ +

ρ 
ρ2 
1
M e (q)ev = − J1( ρ )
M d (q)ev =
ρ
M f (q)ev =
1
cos ρ 
 sin ρ +
 .
ρ
ρ 
(118)
(119)
(120)
(121)
(122)
(123)
An interesting point to observe is that from Eqs. (70) and (72) we find
M c (q) = M e (q) = 0
(124)
since θ = π / 2 in the xy-plane. However, the evanescent parts of these functions are not
zero, and neither are the traveling parts, Eqs. (116) and (117). So two functions which
are identically zero each split in a traveling and evanescent part with opposite sign.
34
In the xy-plane we have in leading order
M a (q)ev =
cos ρ
≈ M b (q)ev
ρ
sin ρ
M f (q)ev ≈
(125)
(126)
ρ
for ρ large (the Bessel functions are O ( 1/ ρ ). This shows that the evanescent waves
along the xy-plane also have an O ( 1/q) part which survives in the far field. The
evanescent part of the Green’s tensor is in leading order for ρ large
tχ (q)ev ≈ (tI − e
cos ρ
ρeρ )
ρ
(127)
and the Green’s vector is
η(q)ev ≈ e ρ
sin ρ
ρ
.
(128)
It follows from Eqs. (24) and (28) that the unsplit Green’s tensor and vector for large
q are
tχ (q) ≈ (tI − qˆ qˆ ) eiq
q
η(q) ≈ − iqˆ
eiq
.
q
(129)
(130)
In the xy-plane we have q = ρ and qˆ = e ρ , so we find
t
t
Re χ (q) ≈ χ (q)ev
(131)
Re η(q) ≈ η(q)ev .
(132)
35
So, in the xy-plane the real parts of the Green’s tensor and vector consist purely of
evanescent waves in the far field, and the imaginary parts are pure traveling. This shows
that in the xy-plane the traveling and evanescent waves contribute “equally” to the far
field. It should also be noted that the evanescent waves here are of the spherical wave
type, unlike on the z-axis. Figure 3 shows a polar graph of M a (q)ev and Re M a (q)tr
for q = 8π , and we see that near the xy-plane the evanescent waves dominate over the
real part of the traveling waves. On the z-axis we have Re M a (q)tr = (cos | z | −1) / | z | ,
and this is zero for q = 8π , so also there the evanescent waves dominate for this value of
q. For other values of q, the traveling and evanescent waves contribute about equally
near the z-axis, but near the xy-plane the evanescent waves dominate over Re M a (q)tr
for all q, since Re M a (q)tr = 0 for all q.
From
Re M a (q)tr + M a (q)ev = Re M a (q) =
cos q
q
(133)
we see that this unsplit function is independent of the polar angle θ . The splitting
introduces a strong angle dependence of both the evanescent part and the real part of the
traveling part, as can be seen in Fig. 3. This angle dependence of the split auxiliary
functions will be studied in detail in the next sections.
36
XIV. Relation to Lommel Functions
The traveling part of M a (q) is given by the integral representation (61) with the upper
limit replaced by α = 1 . Explicitly
1
∫0
M a (q)tr = i dα
α
1−α 2
2
J 0 (αρ ) ei| z | 1−α .
(134)
It turns out that this integral is tabulated (Prudnikov, et.al., 1986b), although the formula
contains a misprint. The factor [exp(ia...)...] should read [-iexp(ia...)...]. The result is
expressed in terms of a Lommel function (Watson, 1922; page 487 of Born and Wolf,
1980), which is a function of two variables, defined as a series with each term containing
a Bessel function. This result has been applied by Bertilone (1991a, 1991b) for the study
of scalar diffraction problems. Since M a (q) = exp(iq ) / q we can obtain the evanescent
part by taking the difference. The result is
M a (q)
ev
∞


m
1
21
=
− tan 2 θ J 2m ( ρ )
J0 (ρ ) + 2

q
m =1


∑(
)
(135)
for | z | > 0 . For | z | < 0 we then use the fact that M a (q)ev is invariant under reflection
in the xy-plane. By taking derivatives as in Sec. IX and with the various relations
between the evanescent parts, given in Sec. X, we can find the other auxiliary functions
in a similar form (Arnoldus and Foley, 2002a). We see from Eq. (135) that M a (q)ev is
expressed in the coordinates ρ and θ , which is a mix of cylinder and spherical
coordinates. For θ = 0 the entire series disappears, and with ρ = 0 for θ = 0 we also
37
have J 0 (0) = 1 , so that M a (q)ev = 1 / q , as in Eq. (107). For larger values of θ all terms
contribute, but the series remains convergent for all θ and ρ . Result (135) is interesting
in its own right, and provides an alternative to numerical integration. However, it does
not shed much light on the behavior of the evanescent waves in the near- and the far field.
Also, the expressions for the other auxiliary functions are not as elegant as Eq. (135), it
seems. In the next section we shall derive our own series expansions, also in terms of
Bessel functions, and the result will be applied to obtain the expansions of M k (q)ev in
series with q as the variable, and in the neighborhood of the origin, e.g., the near field.
The result will exhibit precisely how the evanescent waves determine the near field, and
in particular how the singular behavior at the origin arises.
XV. Expansion in Series with Bessel Functions
In order to arrive at a useful expansion of the evanescent parts of the Green’s tensor and
vector near the origin, we start with the real parts of the traveling parts of the auxiliary
functions. Their integral representations are given in Eqs. (98)-(103). We shall illustrate
the method with Re M a (q)tr and then give the results for the other functions. First we
replace the Bessel function by its series expansion:
∞
J n ( x) =
∑
k =0
(−1) k  x 
 
k!(k + n)!  2 
2k + n
(136)
with n = 0 and x = ρ (1 − u 2 )1 / 2 , and we replace sin(u | z |) by its expansion for small
argument. We then obtain the double series
38
∞
tr
Re M a (q) =
∞
(−1) k + l +1  ρ 
 
(k!) 2 (2l + 1)!  2 
∑∑
k =0 l=0
2k
| z |2l +1
1
∫0
× du (1 − u 2 ) k u 2l +1 .
(137)
The integral on the right-hand side can be evaluated, with result 12 k!l!/(k + l + 1)! . When
we substitute this into Eq. (137) and compare to Eq. (136) then we recognize the
summation over k as the series representation of a Bessel function of order l + 1 . In this
fashion we find the following series representation:
Re M a (q)tr = −
|z|
ρ
∞
∑
l=0
l
l!  2 z 2 
J l +1 ( ρ ) .
−
(2l + 1)!  ρ 
(138)
With | J l +1 ( ρ ) | ≤ 1 and for ρ ≠ 0 , it follows from the ratio test that this series converges.
For ρ → 0 we have to take into account the behavior of the Bessel functions near ρ = 0 ,
given by the first term of the series in Eq. (136). When substituted into the right-hand
side of Eq. (138) it follows again by the ratio test that the series also converges for ρ = 0 .
The series expansions for the other auxiliary functions follow in the same way, with
result:
tr
Re M b (q) =
tr
|z|
ρ
Re M c (q) =
∞
∑
l=0
2
ρ
l
l!  2 z 2 
J l +3 ( ρ )
−
(2l + 1)!  ρ 
∞
∑
l=0
(139)
l
l!  2 z 2 
−
Jl+2 ( ρ )
(2l)!  ρ 
39
(140)
Re M d (q)tr = −
2| z |
ρ2
Re M e (q)tr =
tr
Re M f (q) = −
1
ρ
|z|
ρ
∞
∑
l=0
∞
∑
l=0
∞
∑
l=0
l
(l + 1)!  2 z 2 
Jl+2 ( ρ )
−
(2l + 1)!  ρ 
(141)
l
l!  2 z 2 
J l +1( ρ )
−
(2l)!  ρ 
(142)
l
l!  2 z 2 
−
Jl+2 ( ρ ) .
(2l + 1)!  ρ 
(143)
The most interesting way to look at this is by considering this as Taylor series expansions
in | z | around | z | = 0 for ρ fixed. For | z | = 0 only the first term, l = 0 , contributes,
and we get exactly the result from Sec. XIII, Eqs. (115)-(117). For | z | ≠ 0 we need to
keep more terms. Then, if we calculate Re M k (q)tr with the series expansions above, we
can also find the evanescent parts near the xy-plane with Eq. (96), where Re M k (q) are
the real parts of the right-hand sides of Eqs. (68)-(73). For instance
M a (q)ev =
cos q | z |
1 | z |3
+
J1( ρ ) −
J 2 ( ρ ) + ... .
q
3 ρ2
ρ
(144)
Figure 4 shows M a (q)ev for ρ = 5 , computed this way, and with the series summed
up to l = 20 . It is seen that the series expansion perfectly reproduces the exact result,
obtained with numerical integration, up to about | z | = 12 . If more terms are included, the
range gets larger, but also the computation has to be done in double precision.
The series solution (144) can be seen as an expansion near the xy-plane. On the other
hand, the solution with Lommel functions, Eq. (135), could be considered an expansion
near the z-axis, since for a field point on the z-axis we only need to keep one term. In this
sense, both result are complementary. In the next section we shall derive an expansion
40
which is truly complementary to the above. We shall consider again ρ fixed and | z | as
the variable, but now with | z | large, leading to an asymptotic series in | z | .
XVI. Asymptotic Series
In order to derive an asymptotic expansion for large | z | we start from the integral
representations for M k (q)ev , Eqs. (90)-(95). We notice that these integrals have the
form of Laplace transforms with | z | as the Laplace parameter. The standard procedure
for obtaining an asymptotic expansion for integrals of this type is repeated integration by
parts. In this way we get one term at a time, and every next term becomes more difficult
to obtain. In this section we take a different approach, which leads to the entire
asymptotic series.
As in the previous section, we expand the Bessel function in Eq. (90) in its power
series, Eq. (136), but now we do not expand the exponential. This gives
M a (q)ev =
∞
∑
k =0
(−1) k  ρ 
 
(k!) 2  2 
2k ∞
∫0 du (1 + u 2 )k e−u|z|
(145)
in analogy to Eq. (137). We expand (1 + u 2 ) k with Newton’s binomium and then we
integrate each term. This yields
M a (q)
ev
∞
=
k
∑∑
k =0 l=0
(−1) k (2l)!  ρ 
 
k!l!(k − l)!  2 
2k
1
| z |2l +1
.
Then we change the order of summation and set n = k − l in the summation over k:
41
(146)
M a (q)
ev
∞
=
∞
(−1) n + l (2l)!  ρ 
 
n!l!(n + l)!  2 
∑
∑
l=0 n=0
2 n + 2l
1
| z |2l +1
.
(147)
Here we recognize the summation over n as the series expansion for the Bessel function
J l ( ρ ) , which then gives
M a (q)
ev
1
=
|z|
∞
∑
l=0
l
(2l)!  ρ 
−
 Jl (ρ )
l!  2 z 2 
(148)
in striking resemblance with Eq. (138). For the other auxiliary functions we obtain along
the same lines
M b (q)
M c (q)
ev
M d (q)
ev
=−
ev
M e (q)
M f (q)
1
=−
|z|
| z |2
=−
ev
ev
2
=
∑
l=0
l
∑
l=0
∞
∑
l=0
1
=−
|z|
∞
l
(2l)!  ρ 
−
 J l−2 ( ρ )
l!  2 z 2 
(2l + 1)!  ρ 
−
 J l −1( ρ )
l!  2 z 2 
∞
| z |3
| z |2
∑
l=0
∞
1
1
∞
l
(2l + 2)!  ρ 
−
 Jl (ρ )
l!  2 z 2 
l
(2l + 1)!  ρ 
−
 Jl (ρ )
l!  2 z 2 
∑
l=0
l
(2l)!  ρ 
−
 J l −1( ρ ) .
l!  2 z 2 
(149)
(150)
(151)
(152)
(153)
For Bessel functions with negative order we have J − n ( ρ ) = (−1) n J n ( ρ ) .
For ρ = 0 the only possibly surviving terms are the k = 0 terms, but since J n (0) = 0
for n ≠ 0 , we will only have a non-zero term left if the k = 0 term has the Bessel
function J 0 ( ρ ) . This happens for M a (q)ev , M d (q)ev and M e (q)ev , and the single
42
terms are exactly Eqs. (107)-(109). All others are zero for ρ = 0 , e.g., on the z-axis, in
agreement with Eq. (110). For ρ ≠ 0 the series diverge and they have to be considered
asymptotic series for | z | large, given ρ .
XVII. Evanescent Waves in the Far Field
In the far field, q is large and θ is arbitrary. The standard method to obtain an
asymptotic solution for q large from an angular spectrum representation is by the method
of stationary phase. It is shown in Appendix B that it seems to follow from this method
that only traveling waves contribute to the far field, and it is also shown that this might
not necessarily be true. In any case, we shall consider the contribution of the evanescent
waves to the far field by means of our asymptotic expansion from the previous section.
This asymptotic solution is in terms of the cylinder coordinates ρ and z , so we shall
consider | z | large and ρ arbitrary. Since z = q cosθ , the factors in front of the series
are already O(1/q) or of higher order. For | z | sufficiently large, compared to ρ , at most
the l = 0 term will contribute to the far field. We then find
1
J0 (ρ )
|z|
1
M b (q)ev ≈ −
J2 (ρ )
|z|
1
M f (q)ev ≈
J1( ρ )
|z|
M a (q)ev ≈
(154)
(155)
(156)
and the others are of higher order and therefore give no possible contribution to the far
field.
43
First we notice that on the z-axis we have J 0 (0) = 1 , J1(0) = J 2 (0) = 0 , and Eqs.
(154)-(156) simplify further to
M a (q)ev ≈
1
|z|
(157)
with all others of higher order. The corresponding evanescent parts of the Green’s tensor
and vector are therefore
tχ (q)ev ≈ 1 (tI + e e
z z)
2
1
q
η(q)ev ≈ 0
(158)
(159)
since | z |= q , which is in agreement with Eqs. (111) and (112) up to leading order.
Therefore, the electric field is O(1/q), and the magnetic field does not survive on the zaxis in the far field. Let us now consider ρ large. We can then use the asymptotic form
of the Bessel functions
Jn (ρ ) ≈
2
πρ
cos( ρ − 12 nπ − 14 π ) .
(160)
With ρ = q sin θ we see that the Bessel functions are O(1/ q1 / 2 ), and the three functions
in Eqs. (154)-(156) become O(1/ q3 / 2 ). This shows that M a (q)ev varies from O(1/q) on
the z-axis to O(1/ q3 / 2 ) off the z-axis, and the transition goes smoothly as given by Eq.
(154). The other two functions are zero on the z-axis and they go over in O(1/ q3 / 2 ) off
the z-axis. All other functions remain of higher order. This shows that to leading order
off the z-axis the evanescent waves are O(1/ q3 / 2 ), which drops off faster than O(1/q), and
therefore they do not contribute to the far field. They could be considered to be just in
44
between the far field and the middle field. The transition between O(1/q) and O(1/ q3 / 2 )
occurs where the asymptotic approximation (160) sets in, which is at about ρ = 1 .
Therefore we conclude that there is a cylindrical region around the z-axis with a diameter
of about a wavelength, and inside this cylinder the evanescent waves of the electric field
survive in the far field, whereas outside this cylinder they do not. Since the diameter of
this cylinder is finite, its angular measure ∆θ is zero for q large. So, seen as a function
of θ , the evanescent waves only survive for θ = 0 and π , giving the impression of a
point singularity of no significance, but it should be clear now that such an interpretation
is a consequence of using the wrong coordinates (spherical rather than cylinder
coordinates).
For ρ large it follows from Eq. (160) that J 2 ( ρ ) ≈ − J 0 ( ρ ) , so that M b (q)ev ≈
M a (q)ev . We then find for the Green’s tensor and vector
tχ (q)ev ≈
t
1
1
2
cos(q sin θ − π / 4) ( I − e ρ e ρ )
3 / 2 | cosθ | π sin θ
q
r
η (q)ev ≈
1
1
q 3 / 2 | cosθ |
2
sin( q sin θ − π / 4) e ρ
π sin θ
(161)
(162)
expressed in spherical coordinates. It is interesting to see that also the tensor structure of
tχ (q)ev off the z-axis is different from the tensor structure on the z-axis, as shown in Eq.
(158).
Finally, the asymptotic approximations for the Green’s tensor and vector that hold
both on and off the z-axis, including the smooth transition, are given by
tχ (q)ev ≈ 1 (tI + e e
2
z z ) M a (q)
ev
45
+ 12 (eφ eφ − e ρ e ρ ) M b (q)ev
(163)
η(q) ≈ e ρ M f (q)
(164)
with the auxiliary functions given by (154)-(156). Obviously, the approximation
discussed in this section does not hold near the xy-plane, since we used the asymptotic
expansion for | z | large. However, it is interesting to notice that for a field point in the
xy-plane with ρ large, the same three auxiliary functions have an O(1/q) part, according
to Eqs. (125) and (126). Therefore, the Green’s tensor and vector are identical in form to
Eqs. (163) and (164), but the expressions for the auxiliary functions must be different.
The question now arises whether it would be possible to find expressions for the three
auxiliary functions such that Eqs. (163) and (164) would give the asymptotic (q large, any
θ ) approximation for the Green’s tensor and vector everywhere. This is the topic of the
next section.
XVIII. Uniform Asymptotic Approximation
The behavior of the evanescent waves near the xy-plane follows from Sec. XV, and the
result takes the form as in Eq. (144). The leading term is the total, unsplit, Re M k (q) ,
and the series is a Taylor series in | z | for a fixed ρ . Although this is perfect for
numerical computation, it does not indicate how the solution in the xy-plane goes over in
the typical O(1/ q3 / 2 ) behavior off the xy-plane. In this section we shall derive an
asymptotic approximation which connects the solution in the xy-plane in a smooth way to
the solution off the xy-plane. The method described below was introduced by Berry
(2001) in this problem, who considered the evanescent part of the scalar Green’s
46
function, and this approach was extended by us (Arnoldus and Foley, 2002b) to include
all auxiliary functions of the Green’s tensor. We also improved Berry’s result in that our
solution covers the entire range of angles from the xy-plane up to the z-axis with a single
asymptotic approximation.
A. Derivation
The starting point is the integral representations (90)-(95) for the evanescent parts. It
appears that all six integrals can be covered with one formalism. To this end we write the
integrals in the generic form
∞
∫
M (q)ev = du f (u ) J n ( ρ 1 + u 2 ) e−u| z |
(165)
0
and they differ from each other in the function f (u ) and the order n of the Bessel
function. Table 1 lists f (u ) and n for each of the integrals. Initially, we will be
looking for an asymptotic approximation for M (q)ev in the neighborhood of the xyplane. This implies ρ large, and therefore we can approximate the Bessel function by its
asymptotic approximation, Eq. (160), which we shall now write as
J n ( x) ≈
2
Re (−i )n ei ( x −π / 4) .
πx
(166)
We substitute this into Eq. (165) with x = ρ (1 + u 2 )1 / 2 , and then write the result as
M (q)ev ≈
2
πρ
Re (−i)n e −iπ / 4m(q)
47
(167)
in terms of the new functions m(q) , defined as
∞
∫0
m(q) = du
f (u )
2 1/ 4
(1 + u )
eqw(u ) .
(168)
The complex function w(u ) is
w(u ) = − u | cosθ | + i sin θ 1 + u 2 .
(169)
Equation (168) shows the appearance of the large parameter q in the exponent. We
now wish to make an asymptotic approximation of m(q) for q large, and a given θ . One
critical point of the integrand is the lower limit of integration, u = 0 , and the second one
is the saddle point uo of w(u ) , defined by
w '(uo ) = 0 .
(170)
With Eq. (169) we find that this saddle point is located at
uo = − i | cosθ |
(171)
in the complex u-plane. At the saddle point we have w(uo ) = i . For θ → π / 2 , this
saddle point approaches the lower integration limit, which is also a critical point. We get
the situation that two critical points can be close together. Approximations to this type of
integrals can be made with what is called Bleistein’s method (Olver, 1974; Bleistein and
Handelsman, 1986; Wong, 1989). With Bleistein’s method, we first make a change of
integration variable u → t according to
w(u ) = − u | cosθ | + i sin θ 1 + u 2 = − 12 t 2 + at + b ≡ w(t )
48
(172)
with a and b to be determined. The function w(u ) now goes over in the quadratic form
on the right-hand side. The change of variables also brings the integration curve into the
complex t-plane. We now require that the new curve starts at t = 0 , and that this
corresponds to the beginning of the old curve, u = 0 . We then see immediately that b
must be
b = i sin θ .
(173)
The right-hand side of Eq. (172) now has a saddle point to in the t-plane, which is the
solution of w '(to ) = 0 . We see that to = a , and we now require that under the
transformation the new saddle point is the image of the old saddle point. Since at the
saddle point we have w(uo ) = i , this leads to i = − 12 to2 + ato + β , and with to = a we
then obtain
a = − (1 + i ) 1 − sin θ .
(174)
This saddle point approaches the origin of the t-plane for θ → π / 2 , so we have again
two critical points that approach each other for θ → π / 2 . The contour in the t-plane
follows from the transformation (172), which can be solved for t as a function of u:
t (u ) = − (1 + i ) 1 − sin θ + 2u | cosθ | + 2i (1 − sin θ 1 + u 2 ) .
(175)
For 0 ≤ u < ∞ this then gives the parametrization of the new contour C, which is shown
in Fig. 5 for θ = π / 6 . The integral then becomes
49
∫
m(q) = dt
C
du
f (u )
eqw(t )
2
1
/
4
dt (1 + u )
(176)
with u = u (t ) .
We now approximate the integrand, apart from the exponential, by a linear form
du
f (u )
≈ c1 + c2t
dt (1 + u 2 )1 / 4
(177)
and we choose the constants c1 and c2 such that the approximation is exact in the critical
points t = 0 and t = a . From the transformation (172) we find
du
t−a
=−
dt
w '(u )
(178)
with
w '(u ) = − | cosθ | + i sin θ
u
1+ u
2
.
(179)
Let us first consider the critical point t = 0 , for which u = 0 . Then w '(0) = − | cosθ | ,
and with Eq. (177) with t = 0 we then find c1 = − f (0) a / | cosθ | . Substituting a from
Eq. (174) then gives, after some rearrangements
c1 =
(1 + i ) f (0)
1 + sin θ
(180)
and the values of f (0) for the various functions are given in Table 1. For the second
critical point t = a , u = uo , we have w '(uo ) = 0 and the right-hand side of Eq. (178)
becomes undetermined. From Eq. (175) we have
50
t − a = 2u | cosθ | + 2i (1 − sin θ 1 + u 2 )
(181)
and here the right-hand side has a branch point at u = uo = −i | cosθ | . We expand the
argument of the large square root in a Taylor series around uo , which yields
t−a=
u − uo
− i + ...
sin θ
(182)
and the Taylor expansion of w '(u ) is
w '(u ) = (u − uo )
i
sin 2 θ
+ ... .
(183)
We so obtain
du
= i sin θ .
dt t = a
(184)
Then we set t = a in Eq. (177) and solve for c2 , which gives
c2 =
1
[ f (0) − f (uo )b(θ )]
| cosθ |
(185)
in terms of
b(θ ) = 12 sin θ (1 + sin θ ) .
(186)
The values of f (uo ) are listed in Table 1.
There appears to be a complication with θ → π / 2 , since | cos θ | → 0 in the
denominator on the right-hand side of Eq. (185). But for θ → π / 2 we also have uo → 0
and b(θ ) → 1 , leaving d undetermined. It appears necessary to consider this case as a
51
limit. To this end, we first expand f (uo ) in a Taylor series around u = 0 , as
f (uo ) = f (0) + uo f ' (0) + ... , and then substitute this into Eq. (185), giving
c2 = f (0)
1 − b(θ )
+ if ' (0) + ...
| cosθ |
(187)
where we used uo = −i | cosθ | . The factor (1 − b(θ )) / | cosθ | is still undetermined for
θ → π / 2 . In order to find this limit we expand the numerator and the denominator in a
Taylor series around π / 2 , from which we find
lim
1 − b(θ )
=0
θ →π / 2 | cosθ |
(188)
lim c2 = if '(0) .
(189)
which finally gives
θ →π / 2
The values of f ' (0) are listed in Table 1.
The integrand of the integral in Eq. (176) is analytic for all t, so we can bring the
contour back to the real axis. We then find
m(q) ≈ e
iρ
∞
∫
dt (c1 + c2 t ) e
− 12 qt 2 + aqt
(190)
0
since qb = iρ . This integral can be calculated in closed form. We make the change of
variables ξ = (t − a) q / 2 , which turns the exponent into a perfect square. It also brings
the lower integration limit to ξ = −a q / 2 in the complex ξ -plane. The result can be
expressed in terms of the complementary error function, defined as
52
∞
2
erfc( z ) =
∫
π
2
dξ e−ξ
(191)
z
for z complex. The path of integration runs from z to infinity on the positive real axis.
We then obtain
π
c
m(q) ≈ 2 eiρ + (c1 + a c2 )
q
2q
e
iρ + 12 qa 2
erfc(− a
q
)
2
(192)
With the expressions for c1 , c2 and a this can be simplified further and expressed in
terms of the coordinates as
m(q) ≈ f (uo )
+
π sin θ
ei ( q +π / 4)erfc( i (q − ρ )
2q
1 f (0) − f (uo )b(θ ) iρ
e .
q
| cosθ |
(193)
For later reference we note that the case θ → π / 2 still has to be done with a limit. With
ρ = q , erfc(0) = 1 and Eqs. (185) and (189) we obtain
m(q)
θ =π / 2
≈ f (uo )
π
2q
ei ( q +π / 4) +
i
f ' (0)eiq .
q
(194)
We now substitute the result (193) into Eq. (167) for M (q)ev . After some
rearrangements this yields
M (q)ev ≈
+
f (0) 2
Re (−i )n ei ( ρ −π / 4)
|z| πρ


1
e iρ
Re (−i ) n f (uo ) eiq erfc( i (q − ρ ) ) −
 .
q
iπ (q − ρ ) 

53
(195)
In the first term on the right-hand side we recognize the asymptotic approximation for
J n ( ρ ) from Eq. (166). For reasons explained below we now put this back in. Then we
introduce the function

N (q) = | cosθ | eiρ


1
− eiq erfc( i (q − ρ ) )
iπ ( q − ρ )

(196)
in terms of which the asymptotic approximation becomes
M (q)ev ≈
1
f (0)
Jn (ρ ) −
Re (−i ) n f (uo ) N (q)
|z|
|z|
(197)
and this is the final form. In the definition of N (q ) we have included a factor | cos θ |
which cancels against the same factor in | z | = q | cosθ | in the denominator. The reason
is that in this way the function N (q ) remains finite in the limit θ → π / 2 . To see this we
write N (q ) in the alternative form
N (q) =
1 + sin θ i ( ρ −π / 4)
e
− | cosθ | eiq erfc( i (q − ρ ) )
πq
(198)
from which we have
N (q)
θ =π / 2
=
2 i ( q −π / 4)
e
πq
which is finite.
54
(199)
B. Results
The asymptotic approximation of the evanescent parts of the auxiliary functions M (q)ev
is given by Eq. (197), which involves the universal function N (q ) . Let us temporarily
set
ξ = i (q − ρ ) = (1 + i )
1 q(1 − sin θ ) .
2
(200)
Then N (q ) can be written as

1 −ξ 2 
N (q) = − | cosθ | eiq erfc(ξ ) −
e
 .
ξ π


(201)
For a field point in the xy-plane we have ρ = q , erfc(0) = 1 , and N (q ) is given by Eq.
(199). In particular we see that N (q ) = O ( 1 / q1 / 2 ). On the other hand, off the xy-plane
we have with q − ρ = q (1 − sin θ ) that q − ρ becomes large with q for θ fixed. In that
case, ξ is large and we use the asymptotic approximation for the complementary error
function (Abramowitz and Stegun, 1972)
erfc(ξ ) =
1
ξ π
e
∞


n 1 ⋅ 3 ⋅ ... ⋅ (2n − 1) 
.
1+
(−1)


(2ξ 2 ) n

 n = 1
−ξ 2 
∑
(202)
We see that the first term is just the second term in square brackets in Eq. (201). Since
ξ 2 = i(q − ρ ) the factor exp(−ξ 2 ) does not influence the order, and we find that N (q ) =
O ( 1 / ξ 3 ), which is N (q ) = O ( 1 / q3 / 2 ). Figure 6 shows N (q ) as a function of θ and
for q = 10π . We see indeed that near θ = 90o the real and imaginary parts of N (q )
have a strong peak.
55
To see the structure of result (197) we go back to Eqs. (193) and (194). The second
term on the right-hand sides of both is O ( 1 / q ). The first term on the right-hand side of
Eq. (194) is O ( 1 / q1 / 2 ). Off the xy-plane, we consider Eq. (193) in which erfc (ξ ) =
O ( 1 / q1 / 2 ), making both terms on the right-hand side O ( 1 / q ). It is inherent in
Bleistein’s method that these are the orders that are resolved properly. The next leading
order, which is not resolved, is O ( 1 / q3 / 2 ) (p. 383 of Bleistein and Handelsman, 1986).
To obtain M (q)ev from m(q) , Eq. (167), an additional O ( 1 / q1 / 2 ) appears due to the
1 / ρ . So we see that the leading order of O ( 1 / q ) for M (q)ev in the xy-plane comes
from the term with the complementary error function erfc(ξ ) with ξ = 0 in Eq. (193).
Off the xy-plane both terms become of the same order and both contribute an O ( 1 / q3 / 2 )
to M (q)ev , which is the typical result for the evanescent waves (Sec. XVII).
When the result is written as in Eq. (197), we have to look at this in a different way
because both terms are mixed differently. First of all, due to the 1 / | z | in both terms on
the right-hand side, the case of the xy-plane still has to be considered with a limit. This
factor 1 / | z | is O ( 1 / q ), and for ρ large the Bessel function is O ( 1 / q1 / 2 ), making the
first term the typical O ( 1 / q3 / 2 ). Off the xy-plane, the function N (q ) is O ( 1 / q3 / 2 ), and
this makes the second term O ( 1 / q5 / 2 ), and as indicated in the previous paragraph, this
order is not properly resolved. The fact that this O ( 1 / q5 / 2 ) appears as leading term is a
result of the regrouping of terms in such a way that the second term in brackets in Eq.
(201) is just the leading term of the asymptotic series. Then we might as well drop
N (q ) , and set
M (q)ev ≈
f (0)
Jn (ρ ) .
|z|
56
(203)
Now we compare this to Eqs. (154)-(156) and we see with the values of f (0) and n from
Table 1, that the approximation (203) is the same as the approximation that connected the
value on the z-axis to the field part off the z-axis. In this sense we have made a uniform
asymptotic expansion, which holds for all angles, and reaches the z-axis in the correct
way. This was the reason for putting the Bessel function back in in Eq. (197). When we
now approach the xy-plane, the error function approaches erfc(0) = 1 and N (q ) becomes
O ( 1 / q1 / 2 ). But then it is not clear anymore from Eq. (197) what happens to M (q)ev ,
since this has to be considered with a limit. We go back to Eq. (194) and substitute this
into Eq. (167), which gives for the limit of the xy-plane
M (q)ev ≈

1
2 
Re (−i ) n eiq  f (0) + f ' (0)eiπ / 4

πq 
q

(204)
and this is O ( 1 / q ), provided, of course, that f (0) ≠ 0 .
From the discussion above we see that this O ( 1 / q ) behavior in the xy-plane comes
from erfc (ξ ) ≈ 1 for ξ ≈ 0 . When the argument ξ of the complementary error function
becomes large, so that the asymptotic approximation of erfc(ξ ) sets in, the behavior goes
over in O ( 1 / q3 / 2 ). This happens for | ξ | ≈ 1 , and with Eq. (200) this gives q ≈ 1 + ρ .
With q = ( ρ 2 + z 2 )1 / 2 and | z | << ρ we find by Taylor expansion that this is equivalent
to | z |2 ≈ 2 ρ . So, given ρ , there is a layer of thickness | z | ~ ρ , and within this layer
the evanescent waves are O ( 1 / q ), and end up in the far field. The angular width of this
layer is ∆θ ~| z | / ρ ~ 1 / ρ and this goes to zero for ρ → ∞ , even though the thickness
of the layer grows indefinitely with ρ .
57
With the parameters from Table 1 we find with Eq. (197) the uniform asymptotic
expansions of the auxiliary functions, three of which are
1
[J 0 ( ρ ) − Re N (q)]
|z|
1
M b (q)ev ≈ −
J 2 ( ρ ) + sin 2 θ Re N (q)
|z|
1
[J1( ρ ) − sin θ Im N (q)] .
M f (q)ev ≈
|z|
M a (q)ev ≈
[
]
(205)
(206)
(207)
These are the generalizations of Eqs. (154)-(156). Off the xy-plane the terms with N (q )
are O ( 1 / q5 / 2 ) and are therefore negligible. Then Eqs. (205)-(207) are asymptotically
identical to Eqs. (154)-(156). Near the xy-plane we have to consider this as a limit, which
will be discussed in more detail below. Figures 7 and 8 show the exact M a (q)ev and
its asymptotic approximation (205). We see from Fig. 7 that already for q = 2π the
approximation is excellent, except near the xy-plane. For q = 15π , as shown in Fig. 8,
the approximation near the xy-plane has improved considerably as compared to Fig. 7.
For Fig. 9 we took q = 100π and the exact and approximate solutions are
indistinguishable. This graph also shows that M a (q)ev is much larger near the z-axis
and the xy-plane than in between. This reflects the O ( 1 / q ) and O ( 1 / q3 / 2 ) dependence,
respectively.
The approximations to the other three functions are
2 sin θ
Re N (q)
q
| cosθ |
M d (q)ev ≈ −
Re N (q)
q
1
M e (q)ev ≈ − Im N (q) .
q
M c (q)ev ≈
58
(208)
(209)
(210)
These functions are O ( 1 / q5 / 2 ) off the xy-plane, and are therefore negligible. Near the
xy-plane these functions are O ( 1 / q3 / 2 ), so they do not contribute to the far field along
the xy-plane. Furthermore we see that M d (q )ev is proportional to | cos θ | , which is zero
in the xy-plane, and therefore we can effectively set
M d (q)ev ≈ 0
(211)
everywhere.
Let us now consider the limit θ → π / 2 , for which we use the asymptotic
approximation given by Eq. (204). With the values listed in Table 1 we find for three of
the functions
M a (q)ev ≈ M b (q)ev ≈
M f (q)ev ≈
cos q
q
sin q
.
q
(212)
(213)
The exact values in the xy-plane are given in Sec. XIII. With ρ = q we see from Eq.
(118) that the approximation to M a (q)ev gives the exact value. This seems to be in
contradiction with Figs. 7 and 8 where the exact and asymptotic values near θ = 90o are
not the same. The reason is that the limit for the xy-plane was derived from Eq. (194),
before we replaced the first term on the right-hand side of Eq. (195) with a Bessel
function. Figure 10 shows the asymptotic approximation for M a (q)ev but with J 0 ( ρ )
again replaced with its asymptotic from, Eq. (166). Now we see that that the result near
the xy-plane is indeed exact, but now the approximation diverges near the z-axis. It
seems to be a coincidence that without introducing the Bessel function in Eq. (197) the
result near the xy-plane is better since we already approximated the Bessel function in Eq.
59
(165). Apparently for this case the approximated Bessel function gives a better result
than the exact function.
From Eqs. (119) and (123) we observe that the asymptotic results for M b (q)ev and
M f (q)ev agree to leading order. For M d (q)ev we find the approximate solution
M d (q)ev ≈ 0 in the xy-plane, which agrees with Eq. (121) in leading order. For the
remaining two functions we find
M c (q)ev ≈
2 2
sin(q + π / 4)
q πq
(214)
M e (q)ev ≈
1 2
cos(q + π / 4)
q πq
(215)
and with Eq. (166) we verify that these solutions are asymptotically equivalent to the
exact results from Eqs. (120) and (122).
To conclude this section, let us summarize. As far as the evanescent waves in the far
field, O ( 1 / q ), are concerned, the Green’s tensor and vector are given by Eqs. (163) and
(164), and with the three auxiliary functions given by Eqs. (205)-(207). This accounts for
a far field contribution near the z-axis and near the xy-plane. In between these functions
are O ( 1 / q3 / 2 ), which is also properly resolved. Here, the terms with N (q ) in Eqs.
(205)-(207) are negligible since these are O ( 1 / q5 / 2 ). Then, if one wishes to resolve the
evanescent waves up to O ( 1 / q3 / 2 ) uniformly everywhere, then the terms with M c (q )ev
and M e (q)ev should be added to the Green’s tensor and vector, respectively, since these
functions are O ( 1 / q3 / 2 ) near the xy-plane (and negligible elsewhere). The function
M d (q ) ev never contributes.
60
XIX. Traveling Waves in the Near Field
Let us now turn our attention to the near field. We already found in Sec. VII that the
traveling waves are finite at the origin, and therefore all singular behavior near the origin
must come from the evanescent waves. It turns out to be easier to consider the traveling
waves first. In Sec. XV we obtained series expansions for the functions Re M k (q)tr .
These series are Taylor series in | z | for a given ρ , and the Taylor coefficients became
functions of ρ , involving Bessel functions. We will now be seeking series expansions in
q, around q = 0 , for a given θ . In this way, the Taylor coefficients become functions of
θ.
To this end, we start from the result from Sec. XV, Eqs. (138)-(143). For the Bessel
functions we substitute their series expansion as given by Eq. (136). For Re M a (q)tr we
then find
tr
Re M a (q) = −
|z|
ρ
∞
∞
∑
∑
l=0 k =0
(−1)l + k
l!
ρ
z 2l  
(2l + 1)!k!(k + l + 1)!
2
2k +1
.
(216)
This double series can be written as a single series, similar to the Cauchy product, which
yields
tr
Re M a (q) = −
∞
1
2
n
∑∑
n=0 l=0
(−1) n
l!
ρ
| z |2l +1  
(2l + 1)!(n − l)!(n + 1)!
2
2 n − 2l
. (217)
Then we change variables with ρ = q sin θ , z = q cosθ , and collect the powers of q. We
then obtain the series expansion in q:
61
∞
tr
Re M a (q) = −
1 q | cosθ
2
|
(− 14 q 2 ) n
,
Pn (θ )
∑
n!(n + 1)!
n=0
(218)
with the coefficient functions Pn (θ ) functions of θ . We have split off an overall | cosθ |
for later convenience, and also the factors 1 /(n!(n + 1)!) and 1 / 4n . These overall factors
are taken out in order to keep the coefficient functions Pn (θ ) reasonable, meaning that
they have a very weak n dependence, as will be shown below. If we don’t take out just
the right factors, then the coefficient functions will either increase or decrease very
rapidly with n, leading to numerical problems. The functions Pn (θ ) are explicitly
n
Pn (θ ) = n!
k
(sin 2 θ ) n − k (4 cos2 θ ) k .
∑
(n − k )!(2k + 1)!
k =0
!
(219)
We notice an overall factor n!, which of course cancels against the 1/n! in Eq. (218), but
in this way the function Pn (θ ) becomes well-behaved.
For Re M b (q)tr we find in the same way
tr
Re M b (q) =
∞
1 q 3 sin 2 θ
8
| cosθ |
(− 14 q 2 ) n
,
Pn (θ )
∑
n!(n + 3)!
n=0
(220)
and this series involves the same coefficient function Pn (θ ) . The remaining auxiliary
functions have the series representations
tr
Re M c (q) =
tr
Re M d (q) = −
∞
1 q sin θ
2
Qn (θ )
∑
n!(n + 2)!
n=0
∞
1 q | cosθ
4
|
(− 14 q 2 ) n
(− 14 q 2 ) n
[ Pn (θ ) + Qn (θ )]
∑
n!(n + 2)!
n=0
62
(221)
(222)
tr
Re M e (q) =
tr
Re M f (q) =
∞
1
2
(− 14 q 2 ) n
Qn (θ )
∑
n!(n + 1)!
n=0
− 14 q 2 sin θ
∞
| cosθ |
(− 14 q 2 ) n
Pn (θ )
∑
n!(n + 2)!
n=0
(223)
(224)
and these involve only one more coefficient function, defined by
n
Qn (θ ) = n!
k
(sin 2 θ ) n − k (4 cos 2 θ ) k .
∑
(n − k )!(2k )!
k =0
!
(225)
With Q0 (θ ) = 1 we see that for q = 0 we have Re M e (0)tr = 1 / 2 , and this gives Eq. (58)
for the traveling part of the Green’s vector at the origin. All other functions vanish at the
origin, and the Green’s tensor at the origin only involves the imaginary parts of the
auxiliary functions, and these are pure traveling, giving Eq. (57).
XX. The Coefficient Functions
In order to study the coefficient functions in some detail we introduce the generating
functions
∞
g P (θ ; t ) =
Pn (θ )
tn
n!
(226)
Qn (θ )
tn
.
n!
(227)
∑
n=0
∞
gQ (θ ; t ) =
∑
n=0
63
When we substitute Pn (θ ) from Eq. (219) into the right-hand side of Eq. (226), the result
has the appearance of a Cauchy product of a double series. We then use the Cauchy
product backwards, and express this result in a double series, giving
∞
g P (θ ; t ) =
∞
(t sin 2 θ )l (4t cos 2 θ ) k .
∑
∑
l!(2k + 1)!
k =0 l=0
k!
(228)
Here the summation over l gives an exponential, so that
g P (θ ; t ) = e
t sin 2 θ
∞
∑
k =0
k!
(4t cos 2 θ ) k .
(2k + 1)!
(229)
The series on the right-hand side can be found in a table (Prudnikov, et.al., 1986a), and
we obtain
g P (θ ; t ) =
1
2 cosθ
π
t
et erf ( t cosθ )
(230)
in terms of the error function. Similarly, the generating function for Qn (θ ) is found to be
2
gQ (θ ; t ) = et sin θ + cosθ π t et erf ( t cosθ ) .
(231)
The coefficient functions Pn (θ ) can be recovered from the generating function by
differentiation:
Pn (θ ) =
d n g P (θ ; t )
dt n
64
(232)
t =0
and similarly for Qn (θ ) . To this end, we first expand the error function in its known
series (p. 297 of Abramowitz and Stegun, 1972) and then write the generating function as
∞
g P (θ ; t ) =
∑
k =0
(− cos2 θ ) k k t
(t e ) .
k!(2k + 1)
(233)
Then we differentiate this n times and set t = 0 . This gives
n
Pn (θ ) =
 n  (− cos2 θ ) k
 
 k  2k + 1
k =0
∑
(234)
as an alternative to the form in Eq. (219). With the same procedure for Qn (θ ) the same
sum appears, but with n − 1 , and due to the term exp(t sin 2 θ ) on the right-hand side of
Eq. (231) an extra term (sin θ ) 2n appears. We then find the relation
Qn (θ ) = (sin θ ) 2n + 2n cos2 θ Pn −1(θ ) , n = 1,2,... .
(235)
Then we set sin 2 θ = 1 − cos 2 θ and use Newton’s binomium to represent the nth power,
and we combine the two terms, which gives
n
 n  (− cos 2 θ ) k
 
Qn (θ ) = −
.
k  2k − 1

k =0
∑
(236)
When we set θ = 0 (or π) in Eq. (219) only the term k = n contributes and we find
Pn (0) =
4n (n!) 2
(2n + 1)!
and similarly from Eq. (225)
65
(237)
Qn (0) =
4n (n!) 2
.
(2n)!
(238)
For θ = π / 2 we look at Eqs. (234) and (236) and we see that only the k = 0 contributes,
so that for points in the xy-plane we have
Pn (π / 2) = Qn (π / 2) = 1 .
(239)
To see the behavior of Pn (0) and Qn (0) for n large, we use Stirling’s formula (Arfken
and Weber, 1995) to approximate the factorials. We then find
1 π
2 n
(240)
Qn (0) ≈ π n
(241)
Pn (0) ≈
for n large. This shows that Pn (0) and Qn (0) have a very mild n dependence. It can be
shown (proof omitted here) that Pn (θ ) and Qn (θ ) are bounded by
0 < Pn (θ ) ≤ 1
(242)
1 ≤ Qn (θ ) ≤ 1 + 2n .
(243)
Figure 11 shows the coefficient functions for n = 3 .
For the summation of the series for Re M k (q)tr we need a large number of
coefficient functions, and using Eqs. (234) and (236) repeatedly is not convenient. We
shall now derive recursion relations for the coefficient functions, which will provide a
very efficient method for obtaining these functions. In view of Eq. (234) we temporarily
set
66
n
Yn ( x) =
 n  (−1) k 2k +1
 
x
, n = 0,1,...
k  2k + 1

k =0
∑
(244)
so that Yn (cosθ ) = cosθ Pn (θ ) . Differentiating gives
dYn
= (1 − x 2 ) n .
dx
(245)
From Eq. (244) we see that Yn (0) = 0 , and therefore
x
∫0
Yn ( x) = dt (1 − t 2 ) n .
(246)
Then we write (1 − t 2 ) n = (1 − t 2 )(1 − t 2 ) n −1 , which gives
x
∫0
Yn ( x) = Yn −1 ( x) − dt t 2 (1 − t 2 ) n −1 .
(247)
In this integral we set u (t ) = (1 − t 2 ) n , from which (1 − t 2 ) n −1 = (−2nt ) −1 du / dt .
Integration by parts then gives a relation between Yn (x) and Yn −1( x) , and when we
substitute x = cosθ we obtain the recursion relation for the coefficient functions
(2n + 1) Pn (θ ) = 2nPn −1(θ ) + (sin 2 θ ) n ,
n = 1,2,... .
(248)
This recursion allows us to generate these functions very efficiently from the initial value
of P0 (θ ) = 1 . Then Eq. (235) shows that the Qn (θ ) ’s are related to the Pn (θ ) ’s, and
therefore Eq. (248) implies a recursion relation for the Qn (θ ) ’s. We find this relation to
be
(2n − 1)Qn (θ ) = 2nQn −1(θ ) − (sin 2 θ ) n ,
67
n = 1,2,...
(249)
and the initial value is Q0 (θ ) = 1 . Figure 12 shows the result of a series summation.
XXI. Integral Representations
From Eq. (246), with x = cosθ and Yn (cosθ ) = cosθ Pn (θ ) we find the representation
1
Pn (θ ) =
cosθ
cos θ
∫0 dt (1 − t )
2 n
.
(250)
In this section we shall restrict our attention to z ≥ 0 in order to simplify some of the
notation. We already know that Pn (θ ) and Qn (θ ) are invariant under reflection in the
xy-plane (they only depend on θ through cos 2 θ , Eqs. (234) and (236)), so this is no
limitation. We make the substitution t = cosα in the integrand of Eq. (250), which
yields the alternative representation
1
Pn (θ ) =
cosθ
π /2
∫
θ
dα (sin α ) 2n +1 .
(251)
We now substitute this representation in the series expansion (218). It then appears that
the summation over n can be written as a Bessel function, according to Eq. (136). We
thus obtain the remarkable result (Arnoldus and Foley, 2003c)
π /2
tr
Re M a (q) = −
∫θ dα J1(q sin α ) .
68
(252)
The original representation, Eq. (98), is considerably more complicated in appearance.
We also note that both representations involve a Bessel function of different order. In the
same way we find new representations for the other two auxiliary functions which were
expressed in Pn (θ )
π /2
Re M b (q)tr = sin 2 θ
∫
θ
dα
π /2
∫
θ
tr
Re M f (q) = − sin θ
dα
1
sin 2 α
J 3 (q sin α )
1
J 2 (q sin α ) .
sin α
(253)
(254)
The remaining three functions involve the coefficient functions Qn (θ ) . In order to
obtain interesting integral representations for these functions, we introduce temporarily
the function
n
Z n ( x) =
 n  (−1) k 2k −1
 
x
k  2k − 1

k =1
∑
(255)
in analogy to the function Yn (x) in Eq. (244). It then follows from Eq. (236) that
Qn (θ ) = 1 − cosθ Z n (cosθ ) . Upon differentiating we find that Z n (x) satisfies the
differential equation
[
]
d Zn
1
= 2 (1 − x 2 ) n − 1 .
dx
x
(256)
With Z n (0) = 0 we can integrate this again, as in Eq. (246), and then we set x = cosθ
which yields
69
cos θ
∫0 dt t 2 [1 − (1 − t 2 ) ] .
Qn (θ ) = 1 + cosθ
1
n
(257)
It should be noted that this integral can not be split in two integrals, since both would not
exist in the lower limit. Now we set again t = cosα , which gives
π /2
Qn (θ ) = 1 + cosθ
∫
θ
dα
[1 − (sinα ) ]
cos α
sin α
2n
(258)
2
in analogy to Eq. (251) for Pn (θ ) .
We substitute this representation for Qn (θ ) into Eq. (221). The summation over n
leads again to Bessel functions and we find
Re M c (q)tr =
2
sin θ J 2 (q)
q
1
− sin 2θ
q
π /2
∫
θ
dα
[ J (q sin α ) − sin α
cos α sin α
1
2
2
2
]
J 2 (q ) . (259)
In the same way we find from Eq. (223)
Re M e (q)tr =
1
J1(q)
q
1
− cosθ
q
π /2
1
∫θ dα cos2α [ J1(q sin α ) − sinα J1(q)] .
(260)
Finally, for Re M d (q)tr , Eq. (222), we need the integral representations for both Pn (θ )
and Qn (θ ) , and we obtain
1
Re M d (q)tr = − cosθ J 2 (q)
q
70
1
−
q
π /2
∫
θ
dα
(
)

1 
cos 2 θ
+
J
(
q
sin
α
)
sin 2α J 2 (q ) − J 2 (q sin α )  . (261)
 2
sinα 
cos 2 α

We have verified by numerical integration that these new representations do indeed give
the same results as the old ones.
XXII. Evanescent Waves in the Near Field
Now that we have the real part of the traveling part, we can obtain the evanescent part
from
M k (q)ev = Re M k (q) − Re M k (q)tr .
(262)
The functions M k (q) are given by Eqs. (68)-(73), from which we take the real parts.
These are all standard functions, and we expand these in series around q = 0 . For
instance,
cos q 1
Re M a (q) =
= +
q
q
∞
∑= 0
n
(−1) n +1 q 2n +1
(2n + 2)!
(263)
with the remaining ones similar, but more complicated. We split off the singular terms,
as the 1/q in Eq. (263), and we combine the remaining Taylor series with the series of
Sec. XIX. For instance, we subtract the right-hand side of Eq. (218) from the right-hand
side of Eq. (263). After some serious regrouping, terms with factorials appear, which
have exactly the form of either Eq. (237) or Eq. (238) for Pn (0) and Qn (0) , respectively.
For M a (q)ev we find
71
M a (q)
ev
1
= − 12 q
q
∞
∑= 0
pn (θ )
n
(− 14 q 2 ) n
(264)
n!(n + 1)!
which involves a new coefficient function pn (θ ) , defined in terms of Pn (θ ) as
pn (θ ) = Pn (0) − | cosθ | Pn (θ ) .
(265)
The other functions follow in the same way, for which we need one more coefficient
function
qn (θ ) = | cosθ | Qn (0) − Qn (θ ) .
(266)
The result is
M b (q)
ev
(− 14 q 2 ) n
 3
 1 3 2 ∞
1
q
pn (θ )
= − sin θ  3 +
+  + 8 q sin θ
q

2
q
8
n!(n + 3)!


n=0
(267)
∞
(− 14 q 2 ) n
 3
 1
1
qn (θ )
= | sin 2θ |  3 +  + 2 q sin θ
q

2
q
n!(n + 2)!


n=0
(268)
M c (q)
∑
2
ev
∑
1
1
M d (q)ev = 3 (1 − 3 cos 2 θ ) +
sin 2 θ
2q
q
∞
−
1q
4
(− 14 q 2 ) n
∑= 0[ p (θ )+ | cosθ | q (θ )] n!(n + 2)!
n
n
(269)
n
M e (q)
M f (q)
ev
ev
∞
1
= 2 | cosθ | +
q
1
2
(− 14 q 2 ) n
∑= 0 q (θ ) n!(n + 1)!
n
n
∞
(− 14 q 2 ) n
 1 1 1 2


− q sin θ
pn (θ )
.
= sin θ 2 +
q
2  4
n!(n + 2)!

n=0
∑
72
(270)
(271)
This solution for the evanescent waves in the near field has the remarkable feature
that each “series part” is identical in form, including the overall factor, to the
corresponding solution for Re M k (q)tr from Sec. XIX, under the substitutions
| cosθ | Pn (θ ) → pn (θ )
(272)
Qn (θ ) → qn (θ ) .
(273)
Furthermore, all singular behavior of the evanescent waves appears as additional terms
on the right-hand sides of the equations above. This shows clearly how all singular
behavior of the near field is accounted for by the evanescent waves.
Since pn (θ ) and qn (θ ) are defined in terms of Pn (θ ) and Qn (θ ) , the recursion
relations (248) and (249) imply recursion relations for pn (θ ) and qn (θ ) . We find
(2n + 1) pn (θ ) = 2n pn −1(θ ) − | cosθ | (sin 2 θ ) n
(274)
(2n − 1)qn (θ ) = 2n qn −1(θ ) + (sin 2 θ ) n
(275)
and the initial values are p0 (θ ) = 1− | cosθ | and q0 (θ ) = | cosθ | −1 . The values at θ = 0
(or π ) are
pn (0) = qn (0) = 0
(276)
and for a field point in the xy-plane we obtain
pn (π / 2) = Pn (0) =
4n (n!) 2
(2n + 1)!
qn (π / 2) = − 1 .
73
(277)
(278)
Figure 13 shows these new coefficient functions for n = 3 as a function of θ . The
accuracy of the series expansion is illustrated in Fig. 14, where as an example M b (q)ev
is shown as a function of the radial distance q for θ = 30o .
XXIII. Integral Representations for the Evanescent Waves
In order to obtain new integral representations for the evanescent parts, we need suitable
representations for the functions pn (θ ) and qn (θ ) . We shall assume again that
0 ≤ θ ≤ π / 2 . To this end, we notice that Eq. (245) can also be integrated as
x
∫1
Yn ( x) = Yn (1) + dt (1 − t 2 ) n .
(279)
Since Yn (1) = Pn (0) and Yn (cosθ ) = cosθ Pn (θ ) this can be written as
1


1 
2 n
Pn (θ ) =
Pn (0) −
dt (1 − t ) .

cosθ 


cos θ
∫
(280)
When we compare this to definition (265) of pn (θ ) we find that this is just
1
pn (θ ) =
dt (1 − t )
∫
cos θ
2 n
.
(281)
Similarly, Eq. (256) can be integrated as
x
[
]
1
Z n ( x) = Z n (1) + dt 2 (1 − t 2 ) n − 1 .
t
∫1
74
(282)
With Qn (θ ) = 1 − cosθ Z n (cosθ ) we see that Z n (1) = 1 − Qn (0) . We now split off the “-1
part” in the integrand, which then gives the representation
1


1
2 n

Qn (θ ) = cosθ Qn (0) +
dt 2 (1 − t ) .


t


cos θ
∫
(283)
Comparison with the definition (266) of qn (θ ) then yields
1
qn (θ ) = − cosθ
1
dt 2 (1 − t 2 ) n .
t
∫
cos θ
(284)
For the situation θ → π / 2 we have cosθ → 0 , and the integral does not exist in the
lower limit. However, we can prove the following limit
1
1
lim x dt 2 (1 − t 2 )n = 1
x→ 0
t
∫
(285)
x
which gives qn (π / 2) = −1 , in agreement with Eq. (278).
With the substitution t = cosα in the integrand of Eq. (281) we get
θ
∫0
pn (θ ) = dα (sin α ) 2n +1
(286)
and then we insert this into Eqs. (264), (267) and (271). This yields the new
representations
θ
M a (q)
ev
1
= − dα J1 (q sin α )
q
∫0
75
(287)
M b (q)
ev
 3
1 q 
= − sin 2 θ  3 +
+
+ sin 2 θ
q

2q 8 

θ
θ
1
∫0 dα sin 2 α J3 (q sin α )
 1 1
1
M f (q)ev = sin θ  2 +  − sin θ dα
J 2 (q sin α ) .
q
2 
sin α

0
∫
(288)
(289)
This result has a striking resemblance with Eqs. (252)-(254) for Re M k (q)tr . The
integrals appearing there are the same as here, except that the integration limits are α = θ
and α = π / 2 . When we add for instance Eqs. (252) and (287) we get Re M a (q) for
which we find
1
Re M a (q) = −
q
π /2
∫0 dα J1(q sin α )
(290)
and we know that this is cos q / q . In this form the singular part 1 / q is split off, and this
part is entirely evanescent. The remaining integral is zero at the origin. The remarkable
feature here is that if we split the integration range exactly at the polar angle θ of the
field point q, then the integral over 0 ≤ α ≤ θ represents the remaining evanescent waves
in Re M a (q) , and the integral over θ ≤ α ≤ π / 2 accounts for the traveling waves. The
same conclusion holds for Re M b (q) and Re M f (q) , although for these the parts that are
split off contain a non-singular contribution, equal to − (q / 8) sin 2 θ and (sin θ ) / 2 ,
respectively.
For the series involving qn (θ ) we set t = cosα in the integrand of Eq. (284), giving
θ
∫0
qn (θ ) = − cosθ dα
1
2
cos α
76
(sin α ) 2n +1
(291)
and along similar lines as above we obtain from Eq. (268)
 3
1 
M c (q)ev = sin 2θ  3 + 
q
2q 

θ
1
1
− sin 2θ dα
J 2 (q sin α ) .
2
q
cos α sin α
∫0
(292)
When we compare this to Eq. (259) for Re M c (q)tr we notice that both integrals do not
have the same integrands. With some manipulations, however, we can make them the
same. This gives
 3

2
1 1
M c (q)ev = − sin θ J 2 (q ) + sin 2θ  3 +
+ J 2 (q) 
q

q
2q q


θ
[
]
1
1
− sin 2θ dα
J 2 (q sin α ) − sin 2α J 2 (q ) .
2
q
cos α sin α
∫0
(293)
Here we notice the appearance of − (2 / q ) sin θ J 2 (q ) on the right-hand side. This same
term, but without the minus sign, appeared in Eq. (259) for Re M c (q)tr . When added,
these terms cancel, so we are led to the conclusion that these terms appear due to the
splitting in traveling and evanescent, since in the sum they are absent. This peculiarity
was already observed in Sec. XIII for a field point in the xy-plane, and is now found to
hold more generally. In the form of Eq. (293), also an additional term (sin 2θ ) J 2 (q ) / q
appears on the right-hand side.
In the same way we find
77
M e (q)
ev
θ
1
1
1
= 2 cosθ − cosθ dα
J1(q sin α )
q
q
cos 2 α
∫0
(294)
and in order to make this comparable to Eq. (260) we rearrange this as
1
1
M e (q)ev = 2 cosθ + J1(q ) (cosθ − 1)
q
q
θ
1
1
[J1(q sin α ) − sin α J1(q)] .
− cosθ dα
q
cos 2 α
∫
(295)
0
Here we notice the same phenomenon that a term J1 (q ) / q appears in Re M e (q)tr and
the same term appears with a minus sign in M e (q)ev . Finally for M d (q)ev we find
M d (q)ev =
(
1
1
1 1 2
cosθ J 2 (q ) + 3 (1 − 3 cos 2 θ ) +
sin θ − cos 2 θ J 2 (q )
2
q
q
q
θ
(
)
)

1
1 
cos 2 θ
2
−
−
dα
sin
α
J
(
q
)
J
(
q
sin
α
)
 J 2 (q sin α ) +

2
2
q
sinα 
cos 2 α

∫
(296)
0
and here we have canceling terms of ± cosθ J 2 (q ) / q in the traveling and evanescent
parts.
XXIV. Conclusions
Evanescent waves play an important role in near field optics where spatial resolution of a
radiation field on the order of a wavelength is essential. In an angular spectrum
representation, these waves have wave vectors with a parallel component, with respect to
the xy-plane, corresponding to wavelengths that are smaller than the optical wavelength
78
of the radiation, and as such they serve to resolve details of a radiating source on a scale
smaller than a wavelength. In addition, at distances from the source comparable to a
wavelength or smaller, the evanescent waves dominate over the traveling waves in
amplitude. In fact, all singular behavior of a radiation field at short distances is due to the
evanescent waves. We have studied in detail the nature of the evanescent waves at short
distances by means of a series expansion with the radial distance to the (localized) source
as variable. This was accomplished by considering the Green’s tensor of the electric field
and the Green’s vector of the magnetic field, rather than the fields itself. In this fashion,
the spatial structure of the radiation could be studied independent of the details of the
radiating source. A prime example of a localized source is the electric dipole for which
the fields and the Green’s tensor and vector are essentially identical.
It appears also possible that evanescent waves end up in the far field, together with
the traveling waves, and they contribute to the emitted power. For the electric field, this
happens in a cylindrical region around the z-axis, where the diameter of the cylinder is
about an optical wavelength. The evanescent waves also contribute to the far field near
the xy-plane. One envisions a “sheet” with a thickness that grows with distance to the
source, as the square root of the distance, and within this circular sheet the evanescent
waves in the electric field survive in the far field and contribute to the observable power.
As for the evanescent waves in the magnetic field, they do contribute to the far field
along the xy-plane, but they do not survive along the z-axis.
79
References
Abramowitz, M., and Stegun, I. A. (1972). Handbook of Mathematical Functions. New
York: Dover, p. 298.
Arfken, G. B., and Weber, H. J. (1995). Mathematical Methods for Scientists and
Engineers. 4th edition. San Diego, CA: Academic Press, p. 612.
Arnoldus, H. F., and George, T. F. (1991). Phase-conjugated fluorescence. Phys. Rev. A
43, 3675-3689.
Arnoldus, H. F. (2001). Representation of the near-field, middle-field, and far-field
electromagnetic Green’s functions in reciprocal space. J. Opt. Soc. Am. B 18, 547-555.
Arnoldus, H. F., and Foley, J. T. (2002a). Traveling and evanescent parts of the
electromagnetic Green’s tensor. J. Opt. Soc. Am. A 19, 1701-1711.
Arnoldus, H. F., and Foley, J. T. (2002b). Uniform asymptotic approximation of the
evanescent part of the Green’s tensor. Opt. Commun. 207, 7-15.
Arnoldus, H. F. (2003a). Transverse and longitudinal components of the optical self-,
near-, middle- and far-field. J. Mod. Opt. 50, 755-770.
Arnoldus, H. F., and Foley, J. T. (2003b). Spatial separation of the traveling and
evanescent parts of dipole radiation. Opt. Lett. 28, 1299-1301.
Arnoldus, H. F., and Foley, J. T. (2003c). Traveling and evanescent parts of the optical
near field. J. Mod. Opt. 50, 1883-1901.
Arnoldus, H. F., and Foley, J. T. (2003d). Transmission of dipole radiation through
interfaces and the phenomenon of anti-critical angles. J. Opt. Soc. Am. A, submitted.
Berry, M. V. (2001). Asymptotics of evanescence. J. Mod. Opt. 48, 1535-1541.
80
Bertilone, D. C. (1991a). The contributions of homogeneous and evanescent plane waves
to the scalar optical field: exact diffraction formulae. J. Mod. Opt. 38, 865-875.
Bertilone, D. C. (1991b). Wave theory for a converging spherical incident wave in an
infinite-aperture system. J. Mod. Opt. 38, 1531-1536.
Bleistein, N., and Handelsman, R. A. (1986). Asymptotic Expansions of Integrals. New
York: Dover, p. 380.
Born, M., and Wolf, E. (1980). Principles of Optics, 6th edition. Oxford: Pergamon,
Appendix 5.
Carney, P. S., Fischer, D. G., Foley, J. T., Friberg, A. T., Shchegrov, A. V., Visser, T. D.,
and Wolf, E. (2000). Evanescent waves do contribute to the far field. J. Mod. Opt. 47,
757-758.
Cohen-Tannoudji, C., Dupont-Roc, J., and Grynberg, G. (1989). Photons & Atoms. New
York: Wiley, p. 7.
Courjon, D., and Bainier, C. (1994). Near field microscopy and near field optics. Rep.
Prog. Phys. 57, 989-1028.
Courjon, D. (2003). Near Field Microscopy and Near Field Optics. Singapore: World
Scientific.
de Fornel, F. (2001). Evanescent Waves. From Newtonian Optics to Atomic Optics.
Berlin: Springer-Verlag.
Ford, G. W., and Weber, W. H. (1981). Electromagnetic effects on a molecule at a metal
surface. Surf. Sci. 109, 451-481.
Ford, G. W., and Weber, W. H. (1984). Electromagnetic interactions of molecules with
metal surfaces. Phys. Rep. 113, 195-287.
81
Grattan, K. T. V., and Meggitt, B. T. (2000). Optical Fiber Sensor Technology,
Fundamentals. Dordrecht, The Netherlands: Kluwer.
Jackson, J. D. (1975). Classical Electrodynamics, 2nd edition. New York: Wiley, p. 141.
Keller, O. (1996). Local fields in the electrodynamics of mesoscopic media. Phys. Rep.
268, 85-262.
Keller, O. (1999a). Electromagnetic propagators in micro- and mesoscopic optics, in
Computational Studies of New Materials, edited by D. A. Jelski and T. F. George.
Singapore: World Scientific, pp. 375-439.
Keller, O. (1999b). Attached and radiated electromagnetic fields of an electric point
dipole. J. Opt. Soc. Am. B 16, 835-847.
Lakhtakia, A., and Weiglhofer, W. S. (2000). Evanescent plane waves and the far field:
resolution of a controversy. J. Mod. Opt. 47, 759-763.
Lukosz, W., and Kunz, R. E. (1977a). Light emission by magnetic and electric dipoles
close to a plane interface. I. Total radiated power. J. Opt. Soc. Am. 67, 1607-1615.
Lukosz, W., and Kunz, R. E. (1977b). Light emission by magnetic and electric dipoles
close to a plane interface. II. Radiation patterns of perpendicular oriented dipoles. J. Opt.
Soc. Am. 67, 1615-1619.
Ohtsu, M., editor. (1998). Near-Field Nano/Atom Optics and Technology. New York:
Springer-Verlag.
Olver, F. W. J. (1974). Asymptotics and Special Functions. New York: Academic Press,
p. 344.
Paesler, M. A., and Moyer, P. J. (1996). Near Field Optics, Theory, Instrumentation, and
Applications. New York: Wiley.
82
Pohl, D. W. (1991). Scanning near-field optical microscopy (SNOM), in Advances in
Optical and Electron Microscopy, Vol. 12, edited by T. Mulvey and C. J. R. Sheppard.
San Diego, CA: Academic Press, pp. 243-312.
Pohl, D. W., and Courjon, D., editors. (1993). Near Field Optics, Proceedings of the
NATO Advanced Research Workshop on Near Field Optics, NATO Advanced Study
Institute Series, Vol. 242. Dordrecht, The Netherlands: Kluwer.
Prudnikov, A. P., Brychkov, Yu. A., and Marichev, O. I. (1986a). Integrals and Series,
Vol. 1, Elementary Functions. New York: Gordon and Breach, p. 710, formula 5.2.11.7.
Prudnikov, A. P., Brychkov, Yu. A., and Marichev, O. I. (1986b). Integrals and Series,
Vol. 2, Special Functions. New York: Gordon and Breach, p. 188, formula 2.12.10.3.
Setala, T., and Kaivola, M. (1999). Decomposition of the point-dipole field into
homogeneous and evanescent parts. Phys. Rev. E 59, 1200-1206.
Shchegrov, A. V., and Carney, P. S. (1999). Far-field contribution to the electromagnetic
Green’s tensor from evanescent modes. J. Opt. Soc. Am. A 16, 2583-2584.
Sherman, G. C., Stamnes, J. J., and Lalor, É. (1976). Asymptotic approximations to
angular-spectrum representations. J. Math. Phys. 17, 760-776.
Sipe, J. E. (1981). The dipole antenna problem in surface physics: a new approach. Surf.
Sci. 105, 489-504.
Sipe, J. E. (1987). New Green function formalism for surface optics. J. Opt. Soc. Am. B 4,
481-489.
Tai, C. T. (1971). Dyadic Green’s Functions in Electromagnetic Theory. Scranton:
Intext.
83
van Kranendonk, J., and Sipe, J. E. (1977). Foundations of the macroscopic
electromagnetic theory of dielectric media, in Progress in Optics 15, edited by E. Wolf.
Amsterdam: North-Holland, pp. 246-350.
Watson, G. N. (1922). A Treatise on the Theory of Bessel Functions. Cambridge, UK:
Cambridge U. Press, p. 537.
Wolf, E., and Foley, J. T. (1998). Do evanescent waves contribute to the far field?. Opt.
Lett. 23, 16-18.
Wong, R. (1989). Asymptotic Approximations of Integrals. Boston, MA: Academic Press,
p. 410.
Xiao, M. (1996). A study of resolution limit in optical microscopy: near and far field.
Opt. Commun. 132, 403-409.
Xiao, M. (1999). Evanescent fields do contribute to the far field. J. Mod. Opt. 46, 729733.
Xiao, M. (2000). On the evanescent field of dipole. J. Mod. Opt. 47, 765-768.
84
Appendix A
t
The Green’s tensor χ (q) has a delta function which represents the self field, and
according to Eq. (24) this is
tχ (q) = − 4π δ (q)tI + ... .
3
(A1)
On the other hand, in the angular spectrum representation, Eq. (48), a different delta
t
function appears, e.g., χ (q) = − 4π δ (q)e z e z + ... . This one has a different numerical
factor and a different tensor structure. Since both Green’s tensors are the same, there
must be a hidden delta function in the angular spectrum integral on the right-hand side of
Eq. (48). When written in terms of the auxiliary functions as in Eq. (60), this hidden
delta function must be accounted for by the auxiliary functions. In Sec. VIII it was
shown that by simply comparing Eqs. (24) and (60) that this delta function must be in
M d (q) , as given by Eq. (71). In this Appendix we shall show that the integral
representation (64) for M d (q) does indeed contain this delta function.
To this end, we start with the spectral representation of the delta function:
δ (r ) =
1
∫d k e
3
(2π )
3
ik ⋅r
.
(A2)
~
~
The we use cylinder coordinates (k||,φ , k z ) in k space, and integrate over k|| and φ .
This yields
∞
1
δ (r ) =
δ ( z ) dk|| k|| J 0 (k|| ρ )
2π
∫
0
85
(A3)
and then we change to dimensionless variables as in Sec. VII, which gives
A
1
δ (q) =
δ ( z ) dα α J 0 (α ρ ) .
2π
∫
(A4)
0
Here we have kept the upper limit finite, otherwise the integral does not exist, and it is
understood that “in the end” we take A → ∞ . With the relation x J 0 ( x) = ( x J1( x))' the
integral can be evaluated, and we arrive at the representation
δ (q) =
1
A
δ ( z ) J1( Aρ ) ,
2π
ρ
A→∞ .
(A5)
We now consider the evanescent part M d (q)ev of the integral representation in Eq.
(64), which means that we replace the lower limit α = 0 by α = 1 . First we use
x J 0 ( x) = ( x J1( x))' and integrate by parts. For the integrated part we keep the upper
limit finite as in Eq. (A4). We then obtain
M d (q)ev = −
+
1
ρ
A
ρ
2
J1( Aρ ) A2 − 1 e −| z | A −1
∞
∫
dα α J1(αρ )
1
2

d  2
 α − 1 e −| z | α −1  .
dα 

(A6)
With the representation
δ ( z ) = 12 A2 − 1 e −| z |
86
A2 −1
,
A→∞
(A7)
and Eq. (A5), the first term on the right-hand side of Eq. (A6) becomes − 4π δ (q) . Then
we work out the derivative under the integral, after which it appears that this integral can
be expressed as a combination of the evanescent parts of two other integrals. We find
M d (q)ev = − 4π δ (q) +
1
ρ
M f (q)ev −
|z|
M c (q)ev .
2ρ
(A8)
As the next step we consider the representation (62) for M b (q) . We eliminate J 2 (αρ )
in favor of J1(αρ ) and J 0 (αρ ) with J 2 ( x) = − J 0 ( x) + 2 J1( x) / x , and we use
α2 1
= −β .
β
β
(A9)
This gives the relation
M b (q) = M a (q) − M d (q) −
2
ρ
M f (q)
(A10)
which is Eq. (75). In this derivation we did not use the limits of integration, so this holds
for the separate traveling and evanescent parts as well. Finally, we eliminate M f (q)
between Eqs. (A8) and (A10), which yields
M d (q)ev = −

8π
1
|z|
M c (q)ev  .
δ (q) −  M b (q)ev − M a (q)ev +
3
3
ρ

(A11)
This result has the desired delta function on the right-hand side.
If we would have considered M d (q)tr , with the integration range being 0 ≤ α < 1 ,
then the first term on the right-hand side of Eq. (A6) would have been zero. Equation
(A8) would then become
87
M d (q)tr =
1
ρ
M f (q)tr −
|z|
M c (q)tr
2ρ
(A12)
and since Eq. (A10) also holds for the traveling part, the equivalent of Eq. (A11) is

1
|z|
M d (q)tr = −  M b (q)tr − M a (q)tr +
M c (q)tr  .
3
ρ

(A13)
Appendix B
It was shown in Sec. VI that the Green’s tensor and vector have an angular spectrum
representation, given by Eqs. (48) and (49), respectively. These representations were
obtained from the angular spectrum representation, Eq. (42), of the scalar Green’s
function. These representations define a function of r, and we shall use spherical
coordinates (r ,θ ,φ ) . The goal of the method of stationary phase (Appendix III of Born
and Wolf, 1980) is to obtain an expression for r large, with θ and φ fixed. The phase in
these representations is K ⋅ r = k || ⋅ r + iβ | z | , with β defined by Eq. (2). The method of
stationary phase asserts that the main contribution to an angular spectrum integral comes
from a point in the k || -plane where the phase is stationary (e.g., it has a zero gradient).
Let this point be k ||,o . The idea is that away from this point, the waves are more or less
random, leading to destructive interference, whereas near k ||,o the waves are in phase,
leading to constructive interference. The remainder of the integrand is a function of k || ,
and this function is approximated by its value at k ||,o . The phase is then approximated
by Taylor expansion around the stationary point, leading to a Gaussian form. This phase
88
is then integrated in closed form over the entire k || -plane. The stationary point of K ⋅ r
is
k ||,o = ko sin θ e ρ
(B1)
with e ρ the radial unit vector in the xy-plane corresponding to an observation direction
(θ ,φ ) . The approximation is then
i
2π
∫
d 2k ||
1
β
W (k || )eiK ⋅r ≈ W (k ||,o )
eiko r
.
r
(B2)
The result is an outgoing spherical wave of the far-field type, since it is O(1/r). For a
given field point r, we have rˆ = r / r as the unit vector representing the observation
direction. The projection of r̂ onto the xy-plane is sin θ e ρ , and when multiplied by ko
this gives the stationary point k ||,o . This shows that k ||,o is inside the circle k || = ko in
the k || -plane, and therefore corresponds to a traveling wave exp(iK ⋅ r ) of the angular
spectrum. Since the entire contribution seems to come from the stationary point, one
might conclude that the far field only contains traveling waves of the angular spectrum.
It should also be noted that by considering all observation directions r̂ , we cover the
entire inside of the circle k || = ko , so all traveling waves contribute to the far field.
In general, this conclusion is justified and very useful for interpretation. A detected
wave in the far field, although a spherical wave, has its origin in a single plane wave
coming out of the source, when the field is represented by an angular spectrum (Arnoldus
and Foley, 2003b, 2003d). Now let us consider a field point in the xy-plane. Then
θ = π / 2 , and k ||,o = koe ρ . This stationary point is exactly on the circle k || = ko . In the
method of stationary phase, we expand the phase around this point and then integrate
89
over k || . Since the point k ||,o is on the circle, half of its neighboring wave vectors that
contribute are in the evanescent region. One would therefore expect that in the xy-plane
half of the far field comes from evanescent waves. This is indeed the case, as shown in
Sec. XIII. A more subtle complication arises when the field point is on the z-axis. Then
the stationary point is k ||,o = 0 . It turns out that also in this case the integral over k ||
picks up a contribution from the evanescent range in the k || -plane, so that some
evanescent waves end up in the far field on the z-axis, consistent with the results of Sec.
XII. For more details on this we refer to Sherman, et.al. (1976).
For the angular spectrum of the scalar Green’s function, Eq. (42), we have W (k ||) = 1
and with Eq. (B2) this gives for the asymptotic approximation
eik o r
g (r ) ≈
r
(B3)
and this is the exact result for all r (Eq. (10)). For the Green’s tensor from Eq. (48), the
function W (k || ) involves K at the stationary point, for which we need first β at the
stationary point, which is
β o = ko sgn(z ) cosθ .
(B4)
K o = korˆ
(B5)
From this we find
which yields for the asymptotic approximation of the Green’s tensor
tχ (q) ≈ (tI − qˆ qˆ ) eiq
q
90
.
(B6)
This is indeed the O(1/q) part of the Green’s tensor, as seen from Eq. (24). For the
Green’s vector we obtain
eiq
η(q) ≈ − iqˆ
q
in agreement with Eq. (28).
91
(B7)
Table 1. Table of the various parameters that determine the uniform asymptotic
approximations of the evanescent parts of the auxiliary functions.
M a (q )ev
M b (q )ev
M c (q )ev
M d (q )ev
M e (q )ev
M f (q )ev
f (u )
n
1
0
− (1 + u 2 )
2
2u 1 + u 2
1
− u2
0
u
0
1 + u2
1
f (0)
1
−1
0
0
0
1
f (uo )
1
− sin 2 θ
− i | sin 2θ |
cos 2 θ
− i | cosθ |
sin θ
f ' (0)
0
0
2
0
1
0
92
Figure Captions
Figure 1. Schematic illustration of the traveling and evanescent waves in an angular
spectrum. Each wave has a wave vector with a real-valued k || . If the z-component of the
wave vector is also real, then the wave is traveling, as indicated by the wave vectors K on
the left. At opposite sides of the xy-plane, the z-components of the wave vector differs by
a minus sign, and therefore the propagation direction of the wave is as shown in the
diagram. The wave vector K has a discontinuity at the xy-plane. When the z-component
of the wave vector is imaginary, the wave decays in the directions away from the xyplane, as shown on the right, and they travel along the xy-plane with wave vector k || .
Figure 2. Point P is the projection of the field point r on the xy-plane. We take this
point as the origin of the k || -plane, and we take the new x- and y-axes as shown.
Figure 3. Polar diagram of M a (q)ev and Re M a (q )tr for q = 8π . The sum of these
functions is (cos q ) / q , which is independent of the polar angle. The semi-circle is the
reference zero. We see clearly that near the z-axis and the xy-plane the evanescent part is
significant whereas in between the traveling waves dominate.
Figure 4. Graph of M a (q)ev for ρ = 5 and as a function of z . The thick line is the
exact result and the thin line is the approximation with a series of Bessel functions, Eq.
(144), with 22 terms.
93
Figure 5. Contour in the complex t-plane for the integral in Eq. (176) for θ = π / 6 . Point
P is the saddle point t = a , and the curve approaches a line through the saddle point and
under θ / 2 with the real axis. For θ > π / 2 this angle is (π − θ ) / 2.
Figure 6. Curves a and b are the real and imaginary parts of function N (q ) , shown as a
function of θ for q = 10π .
Figure 7. Function M a (q)ev as a function of θ for q = 2π . The thick line is the exact
solution and the thin line is the uniform asymptotic approximation.
Figure 8. Function M a (q)ev as a function of θ for q = 15π . The thick line is the exact
solution and the thin line is the uniform asymptotic approximation. The only difference
between the two is the small deviation near 90o, highlighted with the circle.
Figure 9. Function M a (q)ev as a function of θ for q = 100π . The difference between
the exact solution and the uniform asymptotic approximation can not be seen anymore.
Here we see that M a (q)ev is much more pronounced near θ = 0o and θ = 90 o , which
reflects the fact that in these regions the evanescent waves end up in the far field.
Figure 10. Same as Fig. 7, but with the Bessel function in the asymptotic approximation
replaced by its asymptotic value. The result at θ = 90 o is now exact but the
approximation diverges at θ = 0o .
94
Figure 11. This graph illustrates the typical behavior of the coefficient functions Pn (θ )
and Qn (θ ) as a function of the polar angle θ .
Figure 12. The thick line is Re M a (q)tr as a function of q for θ = 30o , and the thin line
is the approximation by the series from Eq. (218) summed up to n = 20.
Figure 13. Illustration of the functions pn (θ ) and qn (θ ) .
Figure 14. This graph shows the evanescent part of M b (q) for θ = 30o . The thick line
is the exact value, obtained by numerical integration, and the thin line is the
approximation by series expansion, Eq. (267), up to n = 20.
95
K
k||
k||
xy-plane
K
Figure 1
y
Figure 2
~
y
k ||
~x
~
φ
eφ
eρ
ρ
φ
P
x
0.12
z-axis
Figure 3
tr
0.06
ev
0
-0.06
-0.12
xy-plane
0.06
Figure 4
0.04
0.02
0
-0.02
-0.04
0
5
10
15
z
20
Im
complex t-plane
Re
P
C
θ /2
Figure 5
0.1
Figure 6
a
0.05
0
-0.05
b
-0.1
0
30
60
θ
90
0.2
Figure 7
0.1
0
-0.1
0
30
60
θ
90
0.025
Figure 8
0
-0.025
0
30
60
θ
90
0.003
Figure 9
0.002
0.001
0
-0.001
-0.002
0
30
60
θ
90
0.4
Figure 10
0.3
0.2
0.1
0
-0.1
0
30
60
θ
90
Figure 11
3
Q3 (θ )
2
1
P3 (θ )
0
0
90
θ
180
0.5
Figure 12
0
-0.5
-1
0
5
10
15
20
q
25
0.5
Figure 13
p3 (θ )
0
q3 (θ )
-0.5
-1
0
90
θ
180
0
-0.1
Figure 14
-0.2
0
10
20
q
30